Talks
2026
2026
T2F: Forecasting Irregularly Sampled Time Series via Time Features
Consuelo Bridda, Doriana Medić, Lavinia Chiara Tagliabue, Robert Birke, Marco Aldinucci
ai
Time series forecasting under irregular sampling is challenging because temporal information cannot be inferred solely from sample ordering. Although several neural architectures have been specifically developed for irregularly sampled data, the extent to which forecasting performance depends on model complexity rather than temporal representation remains unclear. In this work, we investigate T2F, a feature-based framework for forecasting regular and irregular time series. The forecasting task is reformulated as a supervised learning problem in which temporal information is explicitly encoded through engineered features. In particular, we compare absolute timestamp representations with local time-difference (Δt) encoding, which captures the temporal spacing between observations and between the last observation and the prediction target. T2F is evaluated via an MLP model on synthetic benchmarks and real-world energy consumption data under both regular and irregular sampling conditions. Experimental comparisons include naive baselines, linear models, tree-based methods, and GRU-D, a recurrent architecture designed for irregular time series. Results show that forecasting performance under irregular sampling depends strongly on temporal representation. While absolute timestamps provide useful global temporal context, local Δt encoding consistently improves predictive accuracy, particularly in short-term forecasting. In the considered real-world irregular setting, the MLP with Δt encoding outperforms traditional baselines and GRU-D despite its substantially simpler architecture. Overall, the results suggest that carefully designed temporal representations can make simple feed-forward models highly competitive for moderately irregular short-term forecasting, with no specialized recurrent or continuous-time architectures.
@misc{26:bridda:t2f-ispcs,
title = {{T2F}: Forecasting Irregularly Sampled Time Series via Time Features},
author = {Consuelo Bridda and Doriana Medi{\'c} and Lavinia Chiara Tagliabue and Robert Birke and Marco Aldinucci},
year = 2026,
month = oct,
address = {Brescia, Italy},
url = {https://datacloud.di.unito.it/index.php/s/rnNkS6ZkCTNPoTL},
note = {Presented by Consuelo Bridda on 7 October 2026},
keywords = {ai},
howpublished = {2026 IEEE International Symposium on Precision Clock Synchronization for Measurement, Control, and Communication (ISPCS)},
abstract = {Time series forecasting under irregular sampling is challenging because temporal information cannot be inferred solely from sample ordering. Although several neural architectures have been specifically developed for irregularly sampled data, the extent to which forecasting performance depends on model complexity rather than temporal representation remains unclear. In this work, we investigate T2F, a feature-based framework for forecasting regular and irregular time series. The forecasting task is reformulated as a supervised learning problem in which temporal information is explicitly encoded through engineered features. In particular, we compare absolute timestamp representations with local time-difference (Δt) encoding, which captures the temporal spacing between observations and between the last observation and the prediction target. T2F is evaluated via an MLP model on synthetic benchmarks and real-world energy consumption data under both regular and irregular sampling conditions. Experimental comparisons include naive baselines, linear models, tree-based methods, and GRU-D, a recurrent architecture designed for irregular time series. Results show that forecasting performance under irregular sampling depends strongly on temporal representation. While absolute timestamps provide useful global temporal context, local Δt encoding consistently improves predictive accuracy, particularly in short-term forecasting. In the considered real-world irregular setting, the MLP with Δt encoding outperforms traditional baselines and GRU-D despite its substantially simpler architecture. Overall, the results suggest that carefully designed temporal representations can make simple feed-forward models highly competitive for moderately irregular short-term forecasting, with no specialized recurrent or continuous-time architectures.}
} Admission to the final exam
Lorenzo Brescia
confidential icsc
@misc{26:brescia:phd:admission:final:exam,
address = {Turin, Italy},
author = {Brescia, Lorenzo},
howpublished = {{Presentation of the PhD admission to the final exam}},
keywords = {icsc, confidential},
month = oct,
title = {{Admission to the final exam}},
url = {https://datacloud.di.unito.it/index.php/s/ZmNa9nTjqBXT2nr},
year = 2026,
bdsk-url-1 = {https://datacloud.di.unito.it/index.php/s/ZmNa9nTjqBXT2nr}} TRUE Protection of Scientific Workflows
Lorenzo Brescia
confidential icsc
@misc{26:brescia:phd:talk,
address = {Turin, Italy},
author = {Brescia, Lorenzo},
howpublished = {{Presentation of the PhD advancements}},
keywords = {icsc, confidential},
month = feb,
title = {{TRUE Protection of Scientific Workflows}},
url = {https://datacloud.di.unito.it/index.php/s/5smA8Ke9moQK2W9},
year = 2026,
bdsk-url-1 = {https://datacloud.di.unito.it/index.php/s/5smA8Ke9moQK2W9}} Alpha Parallel Computing Group: la nostra ricerca
Alessia Antelmi, Gianluca Mittone, Giulio Malenza
analytics fl hpc icsc
This presentation introduces the research activities of the Alpha Parallel Computing Group at the University of Turin, affiliated with the ICSC National Research Centre in HPC, Big Data and Quantum Computing. It outlines three main research areas: parallel and cloud computing, federated learning, and large-scale complex systems. In parallel computing, the group covers equation-based simulation and high-performance numerical algorithms—such as the LSQR algorithm for the Gaia mission—fluid dynamics via Navier–Stokes equations for wind energy and motorsport, and energy-aware optimization across heterogeneous HPC architectures. In federated learning, it motivates privacy-preserving distributed learning under regulations like GDPR, presents the PRAISE medical use case, describes FL training/inference and parallel communication topologies (tree, master-worker, peer-to-peer), and demonstrates cross-facility federations with LLaMA-3 across the Leonardo, LUMI, and MeluXina HPC centres. Finally, for large-scale complex systems, it discusses bottom-up agent-based simulation and top-down graph-based modelling, with examples including the Internet and federated social networks such as Mastodon. The talk concludes with the group's expertise, collaborations, and open-source contributions.
@misc{26:unito:school,
abstract = {This presentation introduces the research activities of the Alpha Parallel Computing Group at the University of Turin, affiliated with the ICSC National Research Centre in HPC, Big Data and Quantum Computing. It outlines three main research areas: parallel and cloud computing, federated learning, and large-scale complex systems. In parallel computing, the group covers equation-based simulation and high-performance numerical algorithms---such as the LSQR algorithm for the Gaia mission---fluid dynamics via Navier--Stokes equations for wind energy and motorsport, and energy-aware optimization across heterogeneous HPC architectures. In federated learning, it motivates privacy-preserving distributed learning under regulations like GDPR, presents the PRAISE medical use case, describes FL training/inference and parallel communication topologies (tree, master-worker, peer-to-peer), and demonstrates cross-facility federations with LLaMA-3 across the Leonardo, LUMI, and MeluXina HPC centres. Finally, for large-scale complex systems, it discusses bottom-up agent-based simulation and top-down graph-based modelling, with examples including the Internet and federated social networks such as Mastodon. The talk concludes with the group's expertise, collaborations, and open-source contributions.},
address = {Turin, Italy},
author = {Antelmi, Alessia and Mittone, Gianluca and Malenza, Giulio},
howpublished = {Department of Computer Science - Orientation day for schools},
keywords = {icsc, fl, analytics, hpc},
month = jan,
title = {Alpha Parallel Computing Group: la nostra ricerca},
url = {https://datacloud.di.unito.it/index.php/s/ZHYDXr3Z95tNfWq},
year = 2026,
bdsk-url-1 = {https://datacloud.di.unito.it/index.php/s/ZHYDXr3Z95tNfWq}} Federated Learning for Critical Applications (FLCA) Workshop Opening (AAAI2026)
Bruno Casella, Samuele Fonio, Mirko Polato, Linara Adilova, Michael Kamp
icsc
@misc{26:casella:FLCAopening,
address = {Singapore},
author = {Bruno Casella and Samuele Fonio and Mirko Polato and Linara Adilova and Michael Kamp},
howpublished = {The 40th Annual AAAI Conference on Artificial Intelligence},
keywords = {icsc},
month = jan,
title = {Federated Learning for Critical Applications (FLCA) Workshop Opening (AAAI2026)},
url = {https://datacloud.di.unito.it/index.php/s/ndi9eqLtJk5Npp5},
year = 2026,
bdsk-url-1 = {https://datacloud.di.unito.it/index.php/s/ndi9eqLtJk5Npp5}} Torino 2033
Marco Aldinucci
IT4LIA
@misc{26:torino2033:ai,
address = {Torino},
author = {Marco Aldinucci},
howpublished = {Torino 2033 - Candidatura a Capitale Europea della Cultura},
keywords = {IT4LIA},
month = feb,
note = {Presentation, 24 February 2026},
title = {Torino 2033},
url = {https://datacloud.di.unito.it/index.php/s/CEgdigErzTpnBXy},
year = 2026,
bdsk-url-1 = {https://datacloud.di.unito.it/index.php/s/CEgdigErzTpnBXy}} Streaming Workflows: From Scientific Applications to AI and Back
Marco Aldinucci
IT4LIA
@misc{26:pdp:streamflow,
address = {Cluj-Napoca, Romania},
author = {Marco Aldinucci},
date-modified = {2026-09-19 19:46:34 +0200},
howpublished = {PDP 2026 - 34th Euromicro International Conference on Parallel, Distributed, and Network-Based Processing},
keywords = {IT4LIA},
month = mar,
note = {Keynote talk},
title = {Streaming Workflows: From Scientific Applications to {AI} and Back},
url = {https://datacloud.di.unito.it/index.php/s/dHLPLE5NdACgdH8},
year = 2026,
bdsk-url-1 = {https://datacloud.di.unito.it/index.php/s/dHLPLE5NdACgdH8}} Sustainable HPC4AI
Marco Aldinucci
IT4LIA
@misc{26:aihub:sustainablehpc4ai,
address = {ITC-ILO, United Nations Campus - Torino},
author = {Marco Aldinucci},
howpublished = {Shaping AI Futures Across Africa and Europe - AI Hub for Sustainable Development},
keywords = {IT4LIA},
month = mar,
title = {Sustainable HPC4AI},
url = {https://datacloud.di.unito.it/index.php/s/yGTHXBYKsbNZAo6},
year = 2026,
bdsk-url-1 = {https://datacloud.di.unito.it/index.php/s/yGTHXBYKsbNZAo6}} Exploring workflow solutions on the OpenFOAM framework
Giulio Malenza, Alberto Mulone, Daniele Ibba, Marco Edoardo Santimaria, Lorenzo Bosio, Iacopo Colonnelli, Robert Birke, Marco Aldinucci
icsc
Computational fluid dynamics (CFD) simulations based on OpenFOAM are widely adopted in both academic and industrial settings. However, their execution typically relies on rigid, monolithic bash scripts that limit portability, scalability, and efficient exploitation of heterogeneous computing resources. This work introduces a workflow-oriented execution model for OpenFOAM that enables more flexible simulation orchestration and improved utilization of heterogeneous architectures. As a case study, we reimplement the well-known OpenFOAM Motorbike tutorial—a benchmark for steady incompressible Reynolds-averaged Navier–Stokes (RANS) simulations of airflow around a motorcycle and rider. The original tutorial follows a strictly sequential execution model defined in a single Allrun script, encompassing mesh generation, parallel decomposition, solver execution, and result reconstruction. We translate this bash-based pipeline into the Common Workflow Language (CWL), a widely adopted, implementation-agnostic workflow specification, and execute it using StreamFlow. This workflow management system supports modular execution, flexible orchestration, and improved resource utilization across heterogeneous platforms. The original execution steps—ranging from environment setup and mesh preparation to parallel solver runs and post-processing—are reorganized into discrete CWL workflow components. The workflow is structured into three main stages: (i) prep, responsible for mesh preparation; (ii) decompose, handling mesh decomposition and refinement; and (iii) foam_run, executing potential flow initialization, the main solver, and mesh reconstruction. Each stage is encapsulated in reusable CWL components, enabling clear separation of concerns and simplified workflow reconfiguration. The resulting workflow is executed on CINECA systems using the official OpenFOAM-v2412 distribution deployed via Docker/Singularity. This study demonstrates how traditional OpenFOAM tutorials can be transformed into portable, reproducible, and extensible workflows. By decoupling simulation phases and explicitly exposing orchestration strategies, the proposed approach lays the groundwork for systematic investigation of performance portability and scalability on heterogeneous and distributed computing environments.
@misc{26:malenza:Workflow_OpenFOAM,
abstract = {Computational fluid dynamics (CFD) simulations based on OpenFOAM are widely adopted in both academic and industrial settings. However, their execution typically relies on rigid, monolithic bash scripts that limit portability, scalability, and efficient exploitation of heterogeneous computing resources. This work introduces a workflow-oriented execution model for OpenFOAM that enables more flexible simulation orchestration and improved utilization of heterogeneous architectures. As a case study, we reimplement the well-known OpenFOAM Motorbike tutorial---a benchmark for steady incompressible Reynolds-averaged Navier--Stokes (RANS) simulations of airflow around a motorcycle and rider. The original tutorial follows a strictly sequential execution model defined in a single Allrun script, encompassing mesh generation, parallel decomposition, solver execution, and result reconstruction. We translate this bash-based pipeline into the Common Workflow Language (CWL), a widely adopted, implementation-agnostic workflow specification, and execute it using StreamFlow. This workflow management system supports modular execution, flexible orchestration, and improved resource utilization across heterogeneous platforms. The original execution steps---ranging from environment setup and mesh preparation to parallel solver runs and post-processing---are reorganized into discrete CWL workflow components. The workflow is structured into three main stages: (i) prep, responsible for mesh preparation; (ii) decompose, handling mesh decomposition and refinement; and (iii) foam_run, executing potential flow initialization, the main solver, and mesh reconstruction. Each stage is encapsulated in reusable CWL components, enabling clear separation of concerns and simplified workflow reconfiguration. The resulting workflow is executed on CINECA systems using the official OpenFOAM-v2412 distribution deployed via Docker/Singularity. This study demonstrates how traditional OpenFOAM tutorials can be transformed into portable, reproducible, and extensible workflows. By decoupling simulation phases and explicitly exposing orchestration strategies, the proposed approach lays the groundwork for systematic investigation of performance portability and scalability on heterogeneous and distributed computing environments.},
address = {Bologna, IT},
author = {Giulio Malenza and Alberto Mulone and Daniele Ibba and Marco Edoardo Santimaria and Lorenzo Bosio and Iacopo Colonnelli and Robert Birke and Marco Aldinucci},
date-modified = {2026-09-19 19:46:55 +0200},
keywords = {icsc},
month = mar,
title = {Exploring workflow solutions on the {OpenFOAM} framework},
url = {https://datacloud.di.unito.it/index.php/s/HwXHg7GicwHcx59},
year = 2026,
bdsk-url-1 = {https://datacloud.di.unito.it/index.php/s/HwXHg7GicwHcx59}} HPC, dalla ricerca all'innovazione
Gianluca Mittone, Marco Aldinucci
hpc icsc
This presentation explores HPC from research to innovation and the opportunities for improvement. It begins with an intuitive introduction to sequential and parallel computing, using analogies such as a tiramisù recipe to explain how problems are decomposed into sub-problems solved concurrently. It then discusses the purposes of parallel computing: strong scaling (solving a fixed problem faster) and weak scaling (solving a larger problem in constant time). The presentation highlights the historical and ongoing impact of HPC through examples like the Leonardo supercomputer at CINECA (the second most powerful in the EU and first for AI applications), HPC6 by Eni (among the world's top supercomputers), and the HPC4AI laboratory at the University of Turin (designed in 2017, anticipating the cloud-HPC model needed for GenAI, with 91
@misc{26:unito:school1,
abstract = {This presentation explores HPC from research to innovation and the opportunities for improvement. It begins with an intuitive introduction to sequential and parallel computing, using analogies such as a tiramis{\`u} recipe to explain how problems are decomposed into sub-problems solved concurrently. It then discusses the purposes of parallel computing: strong scaling (solving a fixed problem faster) and weak scaling (solving a larger problem in constant time). The presentation highlights the historical and ongoing impact of HPC through examples like the Leonardo supercomputer at CINECA (the second most powerful in the EU and first for AI applications), HPC6 by Eni (among the world's top supercomputers), and the HPC4AI laboratory at the University of Turin (designed in 2017, anticipating the cloud-HPC model needed for GenAI, with 91% energy efficiency). It showcases scientific breakthroughs enabled by HPC, including black hole simulations for the film Interstellar and AlphaFold's protein folding prediction. The talk emphasizes the inevitable marriage of HPC and AI, illustrating how performance optimization can achieve near-peak efficiency (e.g., matrix multiplication reaching 42% of theoretical peak). It concludes with reflections on building the future through innovation, drawing parallels to Alessandro Volta's battery, and stressing the importance of research looking 5+ years ahead with public funding to maintain skills and drive economic impact.},
address = {Turin, Italy},
author = {Mittone, Gianluca and Aldinucci, Marco},
howpublished = {Department of Computer Science - Orientation day for schools},
keywords = {icsc, hpc},
month = apr,
title = {HPC, dalla ricerca all'innovazione},
url = {https://datacloud.di.unito.it/index.php/s/Lz4zkmxdgMinonk},
year = 2026,
bdsk-url-1 = {https://datacloud.di.unito.it/index.php/s/Lz4zkmxdgMinonk}} How AI Really Scales
Marco Aldinucci
IT4LIA
@misc{26:innotech:aiscaling,
address = {Milano},
author = {Marco Aldinucci},
howpublished = {InnoTech Community - The European House - Ambrosetti, hosted by NTT DATA},
keywords = {IT4LIA},
month = apr,
title = {How AI Really Scales},
url = {https://datacloud.di.unito.it/index.php/s/R3qDWTLMz9KRq2y},
year = 2026,
bdsk-url-1 = {https://datacloud.di.unito.it/index.php/s/R3qDWTLMz9KRq2y}} HPC-AI: Mens Sana in Corpore Sano
Marco Aldinucci
IT4LIA
@misc{26:aibootcamp:hpcai,
address = {Villa Campolieto - Ercolano (NA)},
author = {Marco Aldinucci},
howpublished = {AI Strategy Bootcamp - Sto{\`a}},
keywords = {IT4LIA},
month = apr,
note = {Inspirational Talk},
title = {{HPC-AI}: Mens Sana in Corpore Sano},
url = {https://datacloud.di.unito.it/index.php/s/s6ocCfrY5Xfwj7w},
year = 2026,
bdsk-url-1 = {https://datacloud.di.unito.it/index.php/s/s6ocCfrY5Xfwj7w}} CINECA IT4LIA: The Most Powerful AI Supercomputer in Italy
Daniele Cesarini
IT4LIA
@misc{26:acw:it4lia,
address = {Palazzo Gnudi - Bologna},
author = {Daniele Cesarini},
howpublished = {Advanced Computing Workshop 2026 - Dell Technologies and NVIDIA},
keywords = {IT4LIA},
month = may,
title = {{CINECA IT4LIA}: The Most Powerful AI Supercomputer in Italy},
url = {https://datacloud.di.unito.it/index.php/s/pwCMYZML2JyBYYM},
year = 2026,
bdsk-url-1 = {https://datacloud.di.unito.it/index.php/s/pwCMYZML2JyBYYM}} L'intelligenza artificiale nella progettazione e realizzazione delle grandi opere infrastrutturali
Marco Aldinucci
IT4LIA
@misc{26:polito:infrastrutture,
address = {Politecnico di Torino - Torino},
author = {Marco Aldinucci},
howpublished = {Workshop ENIA - Ente Nazionale per l'Intelligenza Artificiale},
keywords = {IT4LIA},
month = jul,
title = {L'intelligenza artificiale nella progettazione e realizzazione delle grandi opere infrastrutturali},
url = {https://datacloud.di.unito.it/index.php/s/WnsrkqecNzmWDaG},
year = 2026,
bdsk-url-1 = {https://datacloud.di.unito.it/index.php/s/WnsrkqecNzmWDaG}} Accelerating sharded data parallelism at scale with federated learning
Gianluca Mittone, Marco Aldinucci
dyman fl icsc space
This presentation, delivered by Gianluca Mittone and Marco Aldinucci (University of Turin) at the 32nd International European Conference on Parallel and Distributed Computing (Euro-Par 26, Pisa, August 27th, 2026), introduces FL+FSDP and FL+HSDP, hybrid schemes that accelerate sharded data parallelism at scale by integrating federated learning (FL) aggregations. The talk begins by motivating the work: foundation models (FMs) are driving an exponential growth in hardware demands, with model sizes, training datasets, and parallelism scales increasing by orders of magnitude, as shown by compute-trend analyses. It reviews the taxonomy of FM parallelization and distribution techniques, then focuses on Fully Sharded Data Parallelism (FSDP) and Hybrid Sharded Data Parallelism (HSDP), explaining how model shards are reconstructed through AllGather/ReduceScatter and how HSDP adds a replication dimension via AllReduce. The core contribution is relaxing mathematical coherency with sequential training through FL: between FL aggregations, models can diverge, but if data is IID, FL statistically converges as sequential training, with N (aggregation frequency) providing a knob to balance communication efficiency and coherency. The presentation formalizes communication costs for FSDP, HSDP, FL, FL+FSDP, and FL+HSDP, and shows how FL+HSDP models a three-tier communication hierarchy—sharding, replication, and federation groups—that maps naturally onto complex HPC network topologies like DragonFly+, exploiting different bandwidth levels (intra-node NVLink 1.8 TB/s, inter-node InfiniBand 100 GB/s, and inter-group links). Experimental results on Llama3.1-8B training over 150M tokens on 512 A100 GPUs (128 nodes) of the Leonardo supercomputer show that FL+FSDP processes data up to 8.04× faster and achieves 4.48× lower evaluation perplexity than baseline sharded DP methods, without hyperparameter tuning. Key contributions include the hybrid communication scheme, interconnect-aware training, theoretical communication-cost analysis, and empirical validation. Future work includes extending FL as a fifth orthogonal parallelism axis, automated topology-aware configuration, elimination of manual tuning, and forward-recovery fault tolerance using Federation Groups as alive backups.
@misc{26:europar:fldp,
abstract = {This presentation, delivered by Gianluca Mittone and Marco Aldinucci (University of Turin) at the 32nd International European Conference on Parallel and Distributed Computing (Euro-Par 26, Pisa, August 27th, 2026), introduces FL+FSDP and FL+HSDP, hybrid schemes that accelerate sharded data parallelism at scale by integrating federated learning (FL) aggregations. The talk begins by motivating the work: foundation models (FMs) are driving an exponential growth in hardware demands, with model sizes, training datasets, and parallelism scales increasing by orders of magnitude, as shown by compute-trend analyses. It reviews the taxonomy of FM parallelization and distribution techniques, then focuses on Fully Sharded Data Parallelism (FSDP) and Hybrid Sharded Data Parallelism (HSDP), explaining how model shards are reconstructed through AllGather/ReduceScatter and how HSDP adds a replication dimension via AllReduce. The core contribution is relaxing mathematical coherency with sequential training through FL: between FL aggregations, models can diverge, but if data is IID, FL statistically converges as sequential training, with N (aggregation frequency) providing a knob to balance communication efficiency and coherency. The presentation formalizes communication costs for FSDP, HSDP, FL, FL+FSDP, and FL+HSDP, and shows how FL+HSDP models a three-tier communication hierarchy---sharding, replication, and federation groups---that maps naturally onto complex HPC network topologies like DragonFly+, exploiting different bandwidth levels (intra-node NVLink ~1.8 TB/s, inter-node InfiniBand ~100 GB/s, and inter-group links). Experimental results on Llama3.1-8B training over ~150M tokens on 512 A100 GPUs (128 nodes) of the Leonardo supercomputer show that FL+FSDP processes data up to 8.04× faster and achieves 4.48× lower evaluation perplexity than baseline sharded DP methods, without hyperparameter tuning. Key contributions include the hybrid communication scheme, interconnect-aware training, theoretical communication-cost analysis, and empirical validation. Future work includes extending FL as a fifth orthogonal parallelism axis, automated topology-aware configuration, elimination of manual tuning, and forward-recovery fault tolerance using Federation Groups as alive backups.},
address = {Pisa, Italy},
author = {Mittone, Gianluca and Aldinucci, Marco},
howpublished = {32nd International European Conference on Parallel and Distributed Computing (EuroPar26)},
keywords = {icsc, dyman, space, fl},
month = aug,
note = {Best Paper Award},
title = {Accelerating sharded data parallelism at scale with federated learning},
url = {https://datacloud.di.unito.it/index.php/s/CYHFFwStdqDt4Cj},
year = 2026,
bdsk-url-1 = {https://datacloud.di.unito.it/index.php/s/CYHFFwStdqDt4Cj}} Cittadini digitali nell'era dell'IA e del quantum
Marco Aldinucci
@misc{26:internet40:cittadini,
address = {Pisa, Italy},
author = {Aldinucci, Marco},
howpublished = {Internet-40: Dalla prima connessione ai cittadini digitali},
month = may,
note = {Panel 1},
title = {Cittadini digitali nell'era dell'{IA} e del quantum},
url = {https://it.unipi.it/wp-content/uploads/2026/04/40ANNIDINTERNET_LOCANDINANEW.pdf},
year = 2026,
bdsk-url-1 = {https://it.unipi.it/wp-content/uploads/2026/04/40ANNIDINTERNET_LOCANDINANEW.pdf}} Chi controlla la tecnologia? Infrastrutture, dati e software
Marco Aldinucci
@misc{26:csi:innovation,
address = {Torino, Italy},
author = {Aldinucci, Marco},
howpublished = {{CSI Xperience}: L'innovazione che unisce},
month = oct,
note = {Round table},
title = {Chi controlla la tecnologia? Infrastrutture, dati e software},
url = {https://www.csipiemonte.it/it/evento/linnovazione-che-unisce},
year = 2026,
bdsk-url-1 = {https://www.csipiemonte.it/it/evento/linnovazione-che-unisce}} Sustainable HPC4AI: sperimentare per la sostenibilità dei data center
Lavinia Chiara Tagliabue, Marco Aldinucci
@misc{26:rus:sustainablehpc4ai,
address = {{DAMA -- Tecnopolo Data Manifattura Emilia-Romagna}, Italy},
author = {Tagliabue, Lavinia Chiara and Aldinucci, Marco},
howpublished = {Workshop ``Data Center Energy Management''},
month = oct,
title = {Sustainable {HPC4AI}: sperimentare per la sostenibilit{\`a} dei data center},
url = {https://reterus.it/energia/2273-11233/workshop-data-center-energy-management},
year = 2026,
bdsk-url-1 = {https://reterus.it/energia/2273-11233/workshop-data-center-energy-management}} NextGen AI infrastructures: a taxonomy of bottlenecks
Marco Aldinucci
ai hpc hpc4ai icsc
@misc{26:AI:bottlenecks:chemistry,
address = {Torino, Italy},
author = {Marco Aldinucci},
date-added = {2026-09-13 23:09:23 +0200},
date-modified = {2026-09-19 19:50:10 +0200},
howpublished = {Chemistry through the eyes of AI and ML Chemistry through the eyes of AI and ML},
keywords = {ai, hpc, icsc, hpc4ai},
month = {September},
title = {{NextGen AI} infrastructures: a taxonomy of bottlenecks},
url = {https://datacloud.di.unito.it/index.php/s/B8p37LEMaNndk8t},
year = {2026},
bdsk-url-1 = {https://datacloud.di.unito.it/index.php/s/B8p37LEMaNndk8t}} 2025
2025
Presentazione del Cluster “Quantum Technologies” delle Case delle Tecnologie Emergenti
Marco Aldinucci, Marco Pironti
@misc{25:cte:quantum:technologies,
address = {Povo, Trento, Italy},
author = {Aldinucci, Marco and Pironti, Marco},
howpublished = {Quantum e Cloud: Le Esperienze delle Case delle Tecnologie Emergenti},
month = feb,
title = {Presentazione del Cluster ``Quantum Technologies'' delle Case delle Tecnologie Emergenti},
url = {https://www.mimit.gov.it/images/stories/documenti/R06_PROGRAMMA_TRENTO_2025_def.pdf},
year = 2025,
bdsk-url-1 = {https://www.mimit.gov.it/images/stories/documenti/R06_PROGRAMMA_TRENTO_2025_def.pdf}} La scienza aperta in Italia. Stato dell'arte e prospettive future
Marco Aldinucci
@misc{25:stsitalia:openscience,
address = {Milano, Italy},
author = {Aldinucci, Marco},
howpublished = {10th STS Italia Conference: Technoscience for Good},
month = jun,
note = {Round table},
title = {La scienza aperta in Italia. Stato dell'arte e prospettive future},
url = {https://stsitalia.org/conference-2025-2/},
year = 2025,
bdsk-url-1 = {https://stsitalia.org/conference-2025-2/}} Hybrid Workflows in the Heterogeneous Era
Iacopo Colonnelli
capio eupex streamflow
Thanks to their generality, workflow models represent a powerful abstraction for designing complex applications and executing them on large-scale distributed architectures. However, several additional challenges appear when transitioning from cloud/HPC environments to the entire computing continuum. Continuum execution environments are fully distributed and modular, and modules can be heterogeneous and independent of each other. In addition, continuum workflows often rely on multiple intercommunicating agents that form complex micro-services architectures. Different agents deal with different communication and parallelization paradigms: network-based stream processing at the edge and file-based batch processing on HPC facilities. Finally, support for efficient interactive workflows in the continuum remains an open research problem. This webinar delves into these challenges, exploring how current research addresses them with innovatives techniques and tools (i.e., hybrid workflows and StreamFlow, literate workflows and Jupyter Workflow, transparent streaming workflows and CAPIO), and suggesting potential directions for future investigation.
@misc{25:eupex:streamflow,
abstract = {Thanks to their generality, workflow models represent a powerful abstraction for designing complex applications and executing them on large-scale distributed architectures. However, several additional challenges appear when transitioning from cloud/HPC environments to the entire computing continuum. Continuum execution environments are fully distributed and modular, and modules can be heterogeneous and independent of each other. In addition, continuum workflows often rely on multiple intercommunicating agents that form complex micro-services architectures. Different agents deal with different communication and parallelization paradigms: network-based stream processing at the edge and file-based batch processing on HPC facilities. Finally, support for efficient interactive workflows in the continuum remains an open research problem. This webinar delves into these challenges, exploring how current research addresses them with innovatives techniques and tools (i.e., hybrid workflows and StreamFlow, literate workflows and Jupyter Workflow, transparent streaming workflows and CAPIO), and suggesting potential directions for future investigation.},
author = {Iacopo Colonnelli},
keywords = {eupex, streamflow, capio},
month = feb,
title = {Hybrid Workflows in the Heterogeneous Era},
url = {https://datacloud.di.unito.it/index.php/s/Y76mXwCRKJ2CDa3},
year = 2025,
bdsk-url-1 = {https://datacloud.di.unito.it/index.php/s/Y76mXwCRKJ2CDa3}} A Performance Analysis of VM-based Trusted Execution Environments for Confidential Federated Learning
Bruno Casella
icsc
Federated Learning (FL) is a distributed machine learning approach that has emerged as an effective way to address recent privacy concerns. However, FL introduces the need for additional security measures as FL alone is still subject to vulnerabilities such as model and data poisoning and inference attacks. Confidential Computing (CC) is a paradigm that, by leveraging hardware-based trusted execution environments (TEEs), protects the confidentiality and integrity of ML models and data, thus resulting in a powerful ally of FL applications. Typical TEEs offer an application-isolation level but suffer many drawbacks, such as limited available memory and debugging and coding difficulties. The new generation of TEEs offers a virtual machine (VM)-based isolation level, thus reducing the porting effort for existing applications. In this work, we compare the performance of VM-based and application-isolation level TEEs for confidential FL (CFL) applications. In particular, we evaluate the impact of TEEs and additional security mechanisms such as TLS (for securing the communication channel). The results, obtained across three datasets and two deep learning models, demonstrate that the new VM-based TEEs introduce a limited overhead (at most 1.5x), thus paving the way to leverage public and untrusted computing environments, such as HPC facilities or public cloud, without detriment to performance.
@misc{25:casella:TDXPDP,
abstract = {Federated Learning (FL) is a distributed machine learning approach that has emerged as an effective way to address recent privacy concerns. However, FL introduces the need for additional security measures as FL alone is still subject to vulnerabilities such as model and data poisoning and inference attacks. Confidential Computing (CC) is a paradigm that, by leveraging hardware-based trusted execution environments (TEEs), protects the confidentiality and integrity of ML models and data, thus resulting in a powerful ally of FL applications. Typical TEEs offer an application-isolation level but suffer many drawbacks, such as limited available memory and debugging and coding difficulties. The new generation of TEEs offers a virtual machine (VM)-based isolation level, thus reducing the porting effort for existing applications. In this work, we compare the performance of VM-based and application-isolation level TEEs for confidential FL (CFL) applications. In particular, we evaluate the impact of TEEs and additional security mechanisms such as TLS (for securing the communication channel). The results, obtained across three datasets and two deep learning models, demonstrate that the new VM-based TEEs introduce a limited overhead (at most 1.5x), thus paving the way to leverage public and untrusted computing environments, such as HPC facilities or public cloud, without detriment to performance.},
address = {Torino, Italy},
author = {Bruno Casella},
howpublished = {33rd Euromicro International Conference on Parallel, Distributed and Network-based Processing (PDP 2025)},
keywords = {icsc},
month = mar,
title = {A Performance Analysis of VM-based Trusted Execution Environments for Confidential Federated Learning},
url = {https://datacloud.di.unito.it/index.php/s/CFjA4sqYkbw9Bjy},
year = 2025,
bdsk-url-1 = {https://datacloud.di.unito.it/index.php/s/CFjA4sqYkbw9Bjy}} Cross-Platform Full Waveform Inversion (CPFWI)
Iacopo Colonnelli
icsc streamflow
@misc{25:icsc:cpfwi,
address = {Milano, Italy},
author = {Iacopo Colonnelli},
howpublished = {ICSC Spoke 1 Event},
keywords = {icsc, streamflow},
month = mar,
title = {Cross-Platform Full Waveform Inversion (CPFWI)},
url = {https://datacloud.di.unito.it/index.php/s/3kbBHf39RaFCA9X},
year = 2025,
bdsk-url-1 = {https://datacloud.di.unito.it/index.php/s/3kbBHf39RaFCA9X}} Cross-Facility Federated Learning
Gianluca Mittone, Alberto Mulone, Iacopo Colonnelli, Robert Birke, Marco Aldinucci
eupilot fl icsc space streamflow
Big-tech companies pre-train SOTA LLMs on special-purpose, private HPCs, while public research centres lack the resources to compete. We advocate a new take on large model training, e.g., LLMs, called xFFL, which leverages federated learning as an enabling technique to exploit geographically distributed computing power to bridge such digital divide. This work introduces a proof-of-concept federated training of LLaMA-3 8B on three EuroHPC Top500 facilities, proving the viability of leveraging cross-facility publicly available computational power to sustain SOTA LLM workloads.
@misc{25:cineca:xffl,
abstract = {Big-tech companies pre-train SOTA LLMs on special-purpose, private HPCs, while public research centres lack the resources to compete. We advocate a new take on large model training, e.g., LLMs, called xFFL, which leverages federated learning as an enabling technique to exploit geographically distributed computing power to bridge such digital divide. This work introduces a proof-of-concept federated training of LLaMA-3 8B on three EuroHPC Top500 facilities, proving the viability of leveraging cross-facility publicly available computational power to sustain SOTA LLM workloads.},
address = {Bologna, Italy},
author = {Mittone, Gianluca and Mulone, Alberto and Colonnelli, Iacopo and Birke, Robert and Aldinucci, Marco},
howpublished = {AI & HPC User Forums and HPC User Day},
keywords = {icsc, eupilot, space, fl, streamflow},
month = mar,
title = {Cross-Facility Federated Learning},
url = {https://datacloud.di.unito.it/index.php/s/oXoeHLytc9CZ5KR},
year = 2025,
bdsk-url-1 = {https://datacloud.di.unito.it/index.php/s/oXoeHLytc9CZ5KR}} Advancements in Secure Cloud Computing with Trusted Execution Environments
Lorenzo Brescia
confidential icsc
@misc{25:brescia:phd:talk,
address = {Turin, Italy},
author = {Brescia, Lorenzo},
howpublished = {{Presentation of the PhD advancements}},
keywords = {icsc, confidential},
month = mar,
title = {{Advancements in Secure Cloud Computing with Trusted Execution Environments}},
url = {https://datacloud.di.unito.it/index.php/s/xnZY7QxCfg2bS4M},
year = 2025,
bdsk-url-1 = {https://datacloud.di.unito.it/index.php/s/xnZY7QxCfg2bS4M}} HPC4AI e S-HPC4AI
Marco Aldinucci
ai hpc hpc4ai icsc
@misc{25:cineca:unito:hpc4ai,
address = {Bologna, Italy},
author = {Marco Aldinucci},
howpublished = {CINECA--Universit\`a di Torino},
keywords = {hpc4ai, icsc, hpc, ai},
month = mar,
title = {{HPC4AI} e {S-HPC4AI}},
year = 2025} Federated Learning
Gianluca Mittone
dare fl icsc
This slide deck presents a 10-hour immersive course on Federated Learning (FL) fundamentals, delivered by Gianluca Mittone at the Master in High-Performance Computing (MHPC) at SISSA, May 13th–15th, 2026. The course is organized into 10 hours of theoretical lectures and 10 hours of practical exercises. Theoretical lectures cover six modules: (1) Introduction to Federated Learning; (2) Problem Formulation; (3) Practical Algorithmic Design; (4) Evaluation of Federated Learning Systems; (5) Limits of Standard Practices; and (6) Privacy, Robustness, Fairness, Personalization, followed by final remarks. Practical sessions guide students in building an FL simulator in a single Python process, re-implementing it in a distributed way using PyTorch's distributed backend, and experimenting with an off-the-shelf FL framework (Flower, optional). The course is based on "A Field Guide to Federated Optimization" by Wang et al. (2021). Key topics include the mathematical formulation of FL, FedAvg and its generalizations (FedOpt), communication efficiency, client sampling, data and computational heterogeneity, privacy-preserving protocols (SecAgg, differential privacy), robustness against poisoning attacks, fairness, personalization, and system constraints for real-world deployments. The course emphasizes the distinction between cross-device and cross-silo FL settings throughout, providing a comprehensive foundation for researchers and practitioners entering the field.
@misc{26:ictp:fl,
abstract = {This slide deck presents a 10-hour immersive course on Federated Learning (FL) fundamentals, delivered by Gianluca Mittone at the Master in High-Performance Computing (MHPC) at SISSA, May 13th--15th, 2026. The course is organized into 10 hours of theoretical lectures and 10 hours of practical exercises. Theoretical lectures cover six modules: (1) Introduction to Federated Learning; (2) Problem Formulation; (3) Practical Algorithmic Design; (4) Evaluation of Federated Learning Systems; (5) Limits of Standard Practices; and (6) Privacy, Robustness, Fairness, Personalization, followed by final remarks. Practical sessions guide students in building an FL simulator in a single Python process, re-implementing it in a distributed way using PyTorch's distributed backend, and experimenting with an off-the-shelf FL framework (Flower, optional). The course is based on "A Field Guide to Federated Optimization" by Wang et al. (2021). Key topics include the mathematical formulation of FL, FedAvg and its generalizations (FedOpt), communication efficiency, client sampling, data and computational heterogeneity, privacy-preserving protocols (SecAgg, differential privacy), robustness against poisoning attacks, fairness, personalization, and system constraints for real-world deployments. The course emphasizes the distinction between cross-device and cross-silo FL settings throughout, providing a comprehensive foundation for researchers and practitioners entering the field.},
address = {Trieste, Italy},
author = {Mittone, Gianluca},
howpublished = {Master in High-Performance Computing},
keywords = {icsc, dare, fl},
month = may,
title = {Federated Learning},
url = {https://datacloud.di.unito.it/index.php/s/opkdwZBDepa3b4m},
year = 2025,
bdsk-url-1 = {https://datacloud.di.unito.it/index.php/s/opkdwZBDepa3b4m}} A Performance Analysis for Confidential Federated Learning
Bruno Casella, Iacopo Colonnelli, Gianluca Mittone, Robert Birke, Walter Riviera, Antonio Sciarappa, Carlo Cavazzoni, Marco Aldinucci
icsc
Federated Learning (FL) has emerged as a solution to preserve data privacy by keeping the data locally on each participant's device. However, FL alone is still vulnerable to attacks that can cause privacy leaks. Therefore, additional security measures, at the cost of increasing runtimes, become necessary. The Trusted Execution Environment (TEE) approach offers the highest degree of security during execution. However, TEEs suffer from memory limits which prevent safe end-to-end FL training of modern deep models. State-of-the-art approaches limit secure training to selected layers, failing to avert the full spectrum of attacks or adopt layer-wise training affecting model performance. We benchmark the usage of a library OS (LibOS) to run the full, unmodified end-to-end FL training inside the TEE. We extensively evaluate and model the overhead of the different security mechanisms needed to protect the data and model during computation (TEE), communication (TLS), and storage (disk encryption). The obtained results across three datasets and two models demonstrate that LibOSes are a viable way to seamlessly inject security into FL with limited overhead (at most 2x), offering valuable guidance for researchers and developers aiming to apply FL in data-security-focused contexts.
@misc{25:casella:sgx_sfo,
abstract = {Federated Learning (FL) has emerged as a solution to preserve data privacy by keeping the data locally on each participant's device. However, FL alone is still vulnerable to attacks that can cause privacy leaks. Therefore, additional security measures, at the cost of increasing runtimes, become necessary. The Trusted Execution Environment (TEE) approach offers the highest degree of security during execution. However, TEEs suffer from memory limits which prevent safe end-to-end FL training of modern deep models. State-of-the-art approaches limit secure training to selected layers, failing to avert the full spectrum of attacks or adopt layer-wise training affecting model performance. We benchmark the usage of a library OS (LibOS) to run the full, unmodified end-to-end FL training inside the TEE. We extensively evaluate and model the overhead of the different security mechanisms needed to protect the data and model during computation (TEE), communication (TLS), and storage (disk encryption). The obtained results across three datasets and two models demonstrate that LibOSes are a viable way to seamlessly inject security into FL with limited overhead (at most 2x), offering valuable guidance for researchers and developers aiming to apply FL in data-security-focused contexts.},
address = {San Francisco, US},
author = {Bruno Casella and Iacopo Colonnelli and Gianluca Mittone and Robert Birke and Walter Riviera and Antonio Sciarappa and Carlo Cavazzoni and Marco Aldinucci},
howpublished = {IEEE Security and Privacy Workshops (SPW)},
keywords = {icsc},
month = may,
title = {A Performance Analysis for Confidential Federated Learning},
url = {https://datacloud.di.unito.it/index.php/s/pzx75FGgP43YybF},
year = 2025,
bdsk-url-1 = {https://datacloud.di.unito.it/index.php/s/pzx75FGgP43YybF}} FIT - Futuro Innovazione Torinese
Marco Aldinucci
IT4LIA
@misc{25:fit:aifactory,
address = {OGR - Torino},
author = {Marco Aldinucci},
keywords = {IT4LIA},
month = may,
title = {FIT - Futuro Innovazione Torinese},
url = {https://www.eventbrite.com/e/biglietti-fit-futuro-innovazione-torinese-3-appuntamento-1354422487119?utm_experiment=test_share_listing&aff=ebdsshios},
year = 2025,
bdsk-url-1 = {https://www.eventbrite.com/e/biglietti-fit-futuro-innovazione-torinese-3-appuntamento-1354422487119?utm_experiment=test_share_listing&aff=ebdsshios}} Federated Learning for Foundation Models
Gianluca Mittone
dyman fl icsc
Federated Learning (FL) is a distributed machine learning paradigm that enables multiple clients to collaboratively train models while keeping raw data local. This lecture introduces the fundamental principles of FL and examines their application to foundation models, with particular emphasis on privacy, communication efficiency, data and computational heterogeneity, and system-level constraints. FL is presented as an approach for learning across decentralized data sources without directly transferring sensitive information, using focused model updates that are aggregated by a coordinating server. Privacy can be further strengthened through techniques such as differential privacy and secure aggregation. The lecture discusses the main FL deployment settings, distinguishing cross-device learning, involving potentially millions of intermittently available clients, from cross-silo learning, where organizations are generally more stable and highly available. These settings introduce challenges related to non-IID data, heterogeneous computational resources, limited connectivity, client availability, and communication overhead. Federated optimization therefore requires careful coordination between local computation, communication rounds, privacy requirements, and system constraints. Federated Averaging (FedAvg) is presented as a common optimization approach in which locally trained model parameters are periodically aggregated to produce a shared global model. A particular focus is placed on Cross-Facility Federated Learning (xFFL), which extends FL to geographically distributed HPC infrastructures. The approach uses hierarchical orchestration through StreamFlow, HPC-adapting middleware, and containerized PyTorch deployments to address differences in hardware, software environments, and infrastructure policies. Experiments with LLaMA-2 and LLaMA-3 demonstrate the application of federated learning to large language model workloads across HPC facilities. Overall, FL for foundation models is characterized as an interdisciplinary problem combining machine learning, optimization, privacy, security, cryptography, information theory, and distributed systems. The slides emphasize that effective FL systems require co-design of optimization algorithms with privacy, security, communication, computational, and deployment constraints, while careful evaluation is necessary to distinguish simulation results from behavior in real-world federated environments.
@misc{25:asdmsd:fl,
abstract = {Federated Learning (FL) is a distributed machine learning paradigm that enables multiple clients to collaboratively train models while keeping raw data local. This lecture introduces the fundamental principles of FL and examines their application to foundation models, with particular emphasis on privacy, communication efficiency, data and computational heterogeneity, and system-level constraints. FL is presented as an approach for learning across decentralized data sources without directly transferring sensitive information, using focused model updates that are aggregated by a coordinating server. Privacy can be further strengthened through techniques such as differential privacy and secure aggregation. The lecture discusses the main FL deployment settings, distinguishing cross-device learning, involving potentially millions of intermittently available clients, from cross-silo learning, where organizations are generally more stable and highly available. These settings introduce challenges related to non-IID data, heterogeneous computational resources, limited connectivity, client availability, and communication overhead. Federated optimization therefore requires careful coordination between local computation, communication rounds, privacy requirements, and system constraints. Federated Averaging (FedAvg) is presented as a common optimization approach in which locally trained model parameters are periodically aggregated to produce a shared global model. A particular focus is placed on Cross-Facility Federated Learning (xFFL), which extends FL to geographically distributed HPC infrastructures. The approach uses hierarchical orchestration through StreamFlow, HPC-adapting middleware, and containerized PyTorch deployments to address differences in hardware, software environments, and infrastructure policies. Experiments with LLaMA-2 and LLaMA-3 demonstrate the application of federated learning to large language model workloads across HPC facilities. Overall, FL for foundation models is characterized as an interdisciplinary problem combining machine learning, optimization, privacy, security, cryptography, information theory, and distributed systems. The slides emphasize that effective FL systems require co-design of optimization algorithms with privacy, security, communication, computational, and deployment constraints, while careful evaluation is necessary to distinguish simulation results from behavior in real-world federated environments.},
address = {Trieste, Italy},
author = {Mittone, Gianluca},
howpublished = {ICTP Advanced School on Foundation Models for Scientific Discovery},
keywords = {icsc, dyman, fl},
month = jul,
title = {Federated Learning for Foundation Models},
url = {https://datacloud.di.unito.it/index.php/s/5gzSkSgN8nzKz9D},
year = 2025,
bdsk-url-1 = {https://datacloud.di.unito.it/index.php/s/5gzSkSgN8nzKz9D}} End-to-End Confidentiality with SEV-SNP Leveraging In-Memory Storage
Lorenzo Brescia
confidential icsc
@misc{25:brescia:systex,
address = {Venice, Italy},
author = {Lorenzo Brescia},
keywords = {icsc, confidential},
month = jul,
title = {{End-to-End Confidentiality with SEV-SNP Leveraging In-Memory Storage}},
url = {https://datacloud.di.unito.it/index.php/s/R7EM3GCwmHRJJCq},
year = 2025,
bdsk-url-1 = {https://datacloud.di.unito.it/index.php/s/R7EM3GCwmHRJJCq}} Decentralized Time Series Classification with ROCKET features
Bruno Casella, Matthias Jakobs, Marco Aldinucci, Sebastian Buschjaeger
icsc
Time series classification (TSC) is a critical task with applications in various domains, including healthcare, finance, and industrial monitoring. Due to privacy concerns and data regulations, Federated Learning has emerged as a promising approach for learning from distributed time series data without centralizing raw information. However, most FL solutions rely on a client-server architecture, which introduces robustness and confidentiality risks related to the distinguished role of the server, which is a single point of failure and can observe knowledge extracted from clients. To address these challenges, we propose DROCKS, a fully decentralized FL framework for TSC that leverages ROCKET (RandOm Convolutional KErnel Transform) features. In DROCKS, the global model is trained by sequentially traversing a structured path across federation nodes, where each node refines the model and selects the most effective local kernels before passing them to the successor. Extensive experiments on the UCR archive demonstrate that DROCKS outperforms state-of-the-art client-server FL approaches while being more resilient to node failures and malicious attacks.
@misc{25:casella:drocks,
abstract = {Time series classification (TSC) is a critical task with applications in various domains, including healthcare, finance, and industrial monitoring. Due to privacy concerns and data regulations, Federated Learning has emerged as a promising approach for learning from distributed time series data without centralizing raw information. However, most FL solutions rely on a client-server architecture, which introduces robustness and confidentiality risks related to the distinguished role of the server, which is a single point of failure and can observe knowledge extracted from clients. To address these challenges, we propose DROCKS, a fully decentralized FL framework for TSC that leverages ROCKET (RandOm Convolutional KErnel Transform) features. In DROCKS, the global model is trained by sequentially traversing a structured path across federation nodes, where each node refines the model and selects the most effective local kernels before passing them to the successor. Extensive experiments on the UCR archive demonstrate that DROCKS outperforms state-of-the-art client-server FL approaches while being more resilient to node failures and malicious attacks.},
address = {Porto, Portugal},
author = {Bruno Casella and Matthias Jakobs and Marco Aldinucci and Sebastian Buschjaeger},
howpublished = {3rD Workshop on Advancements in Federated Learning (ECML-PKDD)},
keywords = {icsc},
month = sep,
title = {Decentralized Time Series Classification with ROCKET features},
url = {https://datacloud.di.unito.it/index.php/s/Bp7Y5NmDDBwcizH},
year = 2025,
bdsk-url-1 = {https://datacloud.di.unito.it/index.php/s/Bp7Y5NmDDBwcizH}} Fault Tolerance in Large-Scale ML Training through FL: Fourth Periodic Meeting
Gianluca Mittone
eupilot fl icsc space
This fourth periodic meeting presentation (October 28, 2025) reports on simulating full-scale LLM pre-training to study fault tolerance through FL. It introduces discrete-event and agent-based simulation, implemented with SimPy. Each GPU is modeled as an agent with lifecycle stages—Setup, Compute, Checkpoint, and Wait—organized into groups. Simulation parameters include setup time ( 20 min), compute time ( 1 s per sample), checkpoint interval ( 10 min) and checkpoint time ( 1 min), stochastic failure probability ( 1e-5 per time step), and downtime ( 1 h), each with 10
@misc{25:riken:ftfl5,
abstract = {This fourth periodic meeting presentation (October 28, 2025) reports on simulating full-scale LLM pre-training to study fault tolerance through FL. It introduces discrete-event and agent-based simulation, implemented with SimPy. Each GPU is modeled as an agent with lifecycle stages---Setup, Compute, Checkpoint, and Wait---organized into groups. Simulation parameters include setup time (~20 min), compute time (~1 s per sample), checkpoint interval (~10 min) and checkpoint time (~1 min), stochastic failure probability (~1e-5 per time step), and downtime (~1 h), each with 10% Gaussian variance. Preliminary results compare FSDP, FL-FR (forward recovery), and FL-FS (fail-stop) across node counts (4 GPUs per node). Metrics include overall execution time, setup time, checkpoint count, cumulative checkpointing time, failure count, and cumulative failure time. FL-FS generally reduces checkpointing overhead and failure impact at scale. Next steps include modeling setup and checkpoint failures, using real-world failure rates, simulating different convergence rates, estimating optimal checkpoint latency, and moving time steps to minutes.},
address = {Kobe, Japan},
author = {Mittone, Gianluca},
howpublished = {RIKEN Center for Computational Science},
keywords = {icsc, eupilot, space, fl},
month = sep,
title = {Fault Tolerance in Large-Scale ML Training through FL: Fourth Periodic Meeting},
url = {https://datacloud.di.unito.it/index.php/s/mbWsNzAkmWDbdke},
year = 2025,
bdsk-url-1 = {https://datacloud.di.unito.it/index.php/s/mbWsNzAkmWDbdke}} Fault Tolerance in Large-Scale ML Training through FL: Third Periodic Meeting
Gianluca Mittone
eupilot fl icsc space
This third periodic meeting presentation (October 14, 2025) presents a bottom-up approach to fault tolerance in large-scale ML training through FL. It describes the pre-training hardware/software stack—AI algorithms, frameworks, toolkits, and hardware—and how node-level complexity leads to noisy fault manifestations. It reviews failure frequency and cost, using pie charts from GPT-3 and Platform-X/Unicorn, and categorizes faults by severity (SEV1–SEV3). The talk details NCCL socket retry parameters (NCCL_SOCKET_RETRY_CNT, NCCL_SOCKET_RETRY_SLEEP_MSEC) and NCCL communicator shrinking via ncclCommShrink with NCCL_SHRINK_ABORT. It then presents xFFL deployment scenarios for a single sequential client per node and for distributed multi-node clients, showing two recovery approaches when a node crashes: shrink the NCCL communicator to exclude the failed node, or rebuild the distributed backend to exclude the client. If a node rejoins, the backend is rebuilt and other peers act as alive checkpoints. It compares PyTorch init_process_group and NCCL shrink setup times, advocating framework-handled fault tolerance with minimal user code and low impact on training performance.
@misc{25:riken:ftfl4,
abstract = {This third periodic meeting presentation (October 14, 2025) presents a bottom-up approach to fault tolerance in large-scale ML training through FL. It describes the pre-training hardware/software stack---AI algorithms, frameworks, toolkits, and hardware---and how node-level complexity leads to noisy fault manifestations. It reviews failure frequency and cost, using pie charts from GPT-3 and Platform-X/Unicorn, and categorizes faults by severity (SEV1--SEV3). The talk details NCCL socket retry parameters (NCCL_SOCKET_RETRY_CNT, NCCL_SOCKET_RETRY_SLEEP_MSEC) and NCCL communicator shrinking via ncclCommShrink with NCCL_SHRINK_ABORT. It then presents xFFL deployment scenarios for a single sequential client per node and for distributed multi-node clients, showing two recovery approaches when a node crashes: shrink the NCCL communicator to exclude the failed node, or rebuild the distributed backend to exclude the client. If a node rejoins, the backend is rebuilt and other peers act as alive checkpoints. It compares PyTorch init_process_group and NCCL shrink setup times, advocating framework-handled fault tolerance with minimal user code and low impact on training performance.},
address = {Kobe, Japan},
author = {Mittone, Gianluca},
howpublished = {RIKEN Center for Computational Science},
keywords = {icsc, eupilot, space, fl},
month = sep,
title = {Fault Tolerance in Large-Scale ML Training through FL: Third Periodic Meeting},
url = {https://datacloud.di.unito.it/index.php/s/iokaxmMcyy76B46},
year = 2025,
bdsk-url-1 = {https://datacloud.di.unito.it/index.php/s/iokaxmMcyy76B46}} Fault Tolerance in Large-Scale ML Training through FL: Second Periodic Meeting
Gianluca Mittone
eupilot fl icsc space
This second periodic meeting presentation (September 30, 2025) focuses on fault tolerance for large-scale ML training through FL, with emphasis on LLM pre-training. It contrasts pre-training (huge unlabeled corpora, thousands of GPUs, weeks/months, high failure risk) and fine-tuning (small labeled datasets, few GPUs, hours/days, manageable risk). The talk presents xFFL's goal to make pre-training less communication-bound via intra-silo FedAvg, including layer-wise and bucket-wise aggregation. It analyzes failure scenarios using GPT-3 pre-training statistics on Alibaba Cloud, distinguishing transient faults ( 73
@misc{25:riken:ftfl3,
abstract = {This second periodic meeting presentation (September 30, 2025) focuses on fault tolerance for large-scale ML training through FL, with emphasis on LLM pre-training. It contrasts pre-training (huge unlabeled corpora, thousands of GPUs, weeks/months, high failure risk) and fine-tuning (small labeled datasets, few GPUs, hours/days, manageable risk). The talk presents xFFL's goal to make pre-training less communication-bound via intra-silo FedAvg, including layer-wise and bucket-wise aggregation. It analyzes failure scenarios using GPT-3 pre-training statistics on Alibaba Cloud, distinguishing transient faults (~73% of errors) from hardware faults (~27%) and their downtime costs. Current fault-tolerance approaches are reviewed: checkpointing, TorchSnapshot, SpikingBrain/DLRover Flash Checkpoint, Torchrun elasticity, pipeline-parallelism templates (Bamboo/Ooblack), and Gemini. Challenges include FL's extra redundancy, dynamic restart, zero-redundancy checkpointing, and communicator recreation. The proposed approach is framework-handled fault tolerance using NCCL low-level features---ncclCommAbort, ncclCommShrink, and socket retry parameters---to move from a fail-stop model toward retry and autonomous recovery, reserving rollback as a last resort.},
address = {Kobe, Japan},
author = {Mittone, Gianluca},
howpublished = {RIKEN Center for Computational Science},
keywords = {icsc, eupilot, space, fl},
month = sep,
title = {Fault Tolerance in Large-Scale ML Training through FL: Second Periodic Meeting},
url = {https://datacloud.di.unito.it/index.php/s/AWPR5ASwwZt6gJ7},
year = 2025,
bdsk-url-1 = {https://datacloud.di.unito.it/index.php/s/AWPR5ASwwZt6gJ7}} Fault Tolerance in Large-Scale ML Training through FL: First Periodic Meeting
Gianluca Mittone
eupilot fl icsc space
This first periodic meeting presentation, held at RIKEN R-CCS (September 9, 2025), reviews fault tolerance in large-scale machine learning training through federated learning. It outlines the current xFFL implementation, which supports two deployment typologies: cross-facility (inter-cluster) FL using a master-worker topology coordinated by StreamFlow with built-in retry-rollback fault tolerance, and single-facility (intra-cluster) FL using peer-to-peer NCCL all-reduce, which still lacks fault tolerance. The talk discusses checkpointing and rollback strategies for centralized and decentralized FL, including server-side, client-side, individual, and collective checkpointing. It surveys fault-tolerant communication libraries—OpenMPI/MPICH with ULFM, MVAPICH2 checkpoint/restart features, and NCCL communicator abort/shrink capabilities—and covers PyVerbs for RDMA and Torchrun for elastic distributed training. Key open issues include assumptions of a reliable server, fail-stop client models, and the need for more flexible fault-tolerant communication. The presentation sets the stage for integrating fault tolerance directly into xFFL.
@misc{25:riken:ftfl1,
abstract = {This first periodic meeting presentation, held at RIKEN R-CCS (September 9, 2025), reviews fault tolerance in large-scale machine learning training through federated learning. It outlines the current xFFL implementation, which supports two deployment typologies: cross-facility (inter-cluster) FL using a master-worker topology coordinated by StreamFlow with built-in retry-rollback fault tolerance, and single-facility (intra-cluster) FL using peer-to-peer NCCL all-reduce, which still lacks fault tolerance. The talk discusses checkpointing and rollback strategies for centralized and decentralized FL, including server-side, client-side, individual, and collective checkpointing. It surveys fault-tolerant communication libraries---OpenMPI/MPICH with ULFM, MVAPICH2 checkpoint/restart features, and NCCL communicator abort/shrink capabilities---and covers PyVerbs for RDMA and Torchrun for elastic distributed training. Key open issues include assumptions of a reliable server, fail-stop client models, and the need for more flexible fault-tolerant communication. The presentation sets the stage for integrating fault tolerance directly into xFFL.},
address = {Kobe, Japan},
author = {Mittone, Gianluca},
howpublished = {RIKEN Center for Computational Science},
keywords = {icsc, eupilot, space, fl},
month = sep,
title = {Fault Tolerance in Large-Scale ML Training through FL: First Periodic Meeting},
url = {https://datacloud.di.unito.it/index.php/s/W9J2i5PnZs5DiJk},
year = 2025,
bdsk-url-1 = {https://datacloud.di.unito.it/index.php/s/W9J2i5PnZs5DiJk}} Pushing Federated Learning Boundaries: Three Innovative Distributed Intelligence Approaches
Gianluca Mittone
eupilot fl icsc space
This kick-off presentation, delivered by Gianluca Mittone at RIKEN R-CCS (Kobe, Japan, September 2, 2025), introduces three innovative distributed intelligence approaches for pushing federated learning (FL) boundaries. After an overview of the speaker's academic and research background, the talk motivates FL through privacy regulations and the standard FedAvg algorithm. It then presents three main contributions: (1) Model-Agnostic Federated Learning (MAFL/OpenFL-x), which extends OpenFL to support traditional machine learning models via AdaBoost.F, enabling non-deep model federation; (2) High-Performance Federated Learning (FastFL), a FastFlow/RISC-pb2l-based backend supporting multiple communication topologies—master-worker, peer-to-peer, and tree—with strong and weak scaling on HPC and heterogeneous microarchitectures; and (3) Cross-Facility Federated Learning (xFFL), which uses StreamFlow to coordinate federated training across multiple HPC facilities, demonstrated with LLaMA-2/3 pre-training across Leonardo, LUMI, and MeluXina. The talk concludes with future work on RDMA-based fault tolerance and extra applications such as edge-distributed multiview detection and drug-target interaction.
@misc{25:riken:ftfl0,
abstract = {This kick-off presentation, delivered by Gianluca Mittone at RIKEN R-CCS (Kobe, Japan, September 2, 2025), introduces three innovative distributed intelligence approaches for pushing federated learning (FL) boundaries. After an overview of the speaker's academic and research background, the talk motivates FL through privacy regulations and the standard FedAvg algorithm. It then presents three main contributions: (1) Model-Agnostic Federated Learning (MAFL/OpenFL-x), which extends OpenFL to support traditional machine learning models via AdaBoost.F, enabling non-deep model federation; (2) High-Performance Federated Learning (FastFL), a FastFlow/RISC-pb2l-based backend supporting multiple communication topologies---master-worker, peer-to-peer, and tree---with strong and weak scaling on HPC and heterogeneous microarchitectures; and (3) Cross-Facility Federated Learning (xFFL), which uses StreamFlow to coordinate federated training across multiple HPC facilities, demonstrated with LLaMA-2/3 pre-training across Leonardo, LUMI, and MeluXina. The talk concludes with future work on RDMA-based fault tolerance and extra applications such as edge-distributed multiview detection and drug-target interaction.},
address = {Kobe, Japan},
author = {Mittone, Gianluca},
howpublished = {RIKEN Center for Computational Science},
keywords = {icsc, eupilot, space, fl},
month = sep,
title = {Pushing Federated Learning Boundaries: Three Innovative Distributed Intelligence Approaches},
url = {https://datacloud.di.unito.it/index.php/s/Z9ZD2c3MPya2CS3},
year = 2025,
bdsk-url-1 = {https://datacloud.di.unito.it/index.php/s/Z9ZD2c3MPya2CS3}} BookedSlurm: meeting user needs for advanced resource reservations in Slurm
Sandro Gepiro Contaldo, Lorenzo Bosio, Janneth Estefania Hoyos Rea, Elisa Li Perottino, Sergio Rabellino, Marco Aldinucci, Marco Beccuti, Iacopo Colonnelli
icsc
Modern scientific discovery is frequently backed by large-scale scientific experiments, which cannot prescind from the unparalleled computational capabilities provided by high-performance computing systems. However, large data centers usually prioritize system efficiency over user accessibility, posing challenges for researchers without advanced computer science expertise. This work introduces BookedSlurm, a secure and user-focused extension of the Slurm workload manager, aiming to democratize HPC access across interdisciplinary research domains. BookedSlurm enables a partially decentralized regulation of fine-grained advanced resource reservations through a novel credit-based framework, ensuring fair and predictable access to computing resources. Its modular architecture leverages dedicated microservices to manage reservations, credit handling, and accounting. These components are exposed through a secure REST API and an intuitive web-based dashboard, enhancing system usability for novice and expert users. While the dashboard simplifies interactions for non-specialists, advanced users can directly access agent-level APIs for more complex and automated operations. The effectiveness of BookedSlurm is validated on a real-world bioinformatics use case from the SUS-MIRRI.IT project, showcasing its ability to enhance usability, optimize job scheduling, and streamline execution workflows.
@misc{25:escience:bookedslurm,
abstract = {Modern scientific discovery is frequently backed by large-scale scientific experiments, which cannot prescind from the unparalleled computational capabilities provided by high-performance computing systems. However, large data centers usually prioritize system efficiency over user accessibility, posing challenges for researchers without advanced computer science expertise. This work introduces BookedSlurm, a secure and user-focused extension of the Slurm workload manager, aiming to democratize HPC access across interdisciplinary research domains. BookedSlurm enables a partially decentralized regulation of fine-grained advanced resource reservations through a novel credit-based framework, ensuring fair and predictable access to computing resources. Its modular architecture leverages dedicated microservices to manage reservations, credit handling, and accounting. These components are exposed through a secure REST API and an intuitive web-based dashboard, enhancing system usability for novice and expert users. While the dashboard simplifies interactions for non-specialists, advanced users can directly access agent-level APIs for more complex and automated operations. The effectiveness of BookedSlurm is validated on a real-world bioinformatics use case from the SUS-MIRRI.IT project, showcasing its ability to enhance usability, optimize job scheduling, and streamline execution workflows.},
address = {Chicago, IL, USA},
author = {Sandro Gepiro Contaldo and Lorenzo Bosio and Janneth Estefania Hoyos Rea and Elisa Li Perottino and Sergio Rabellino and Marco Aldinucci and Marco Beccuti and Iacopo Colonnelli},
howpublished = {21st IEEE International Conference on e-Science},
keywords = {icsc},
month = sep,
title = {{BookedSlurm}: meeting user needs for advanced resource reservations in {Slurm}},
url = {https://datacloud.di.unito.it/index.php/s/xT58tKdAjJDKy59},
year = 2025,
bdsk-url-1 = {https://datacloud.di.unito.it/index.php/s/xT58tKdAjJDKy59}} Fed2RC: Federated Rocket Kernels and Ridge Classifier for Time Series Classification
Bruno Casella, Samuele Fonio, Lorenzo Sciandra, Claudio Gallicchio, Marco Aldinucci, Mirko Polato, Roberto Esposito
icsc
Time series classification is a pivotal task in modern machine learning, with widespread applications in fields such as healthcare, finance, and cybersecurity. While deep learning methods dominate recent developments, their resource demands and privacy limitations hinder deployment on low-power and decentralized environments. To address these challenges, we introduce Fed2RC, a fully federated and gradient-free approach that integrates the efficiency of Rocket-based feature extraction with the robustness of ridge regression in a privacy-preserving setting. Fed2RC builds upon two key ideas: (i) federated selection and aggregation of high-performing random convolution kernels, and (ii) incremental and communication-efficient updates of ridge classifier parameters using closed-form solutions. Additionally, we propose a novel federated protocol for selecting the global ridge regularization parameter lambda, and show how to improve the communication efficiency by matrix factorization techniques. Extensive experiments on the UCR benchmark demonstrate that Fed2RC achieves state-of-the-art results with a fraction of the computation and communication costs. Code to reproduce the experiments can be found at: https://github.com/CasellaJr/Fed2RC.
@misc{25:casella:Fed2RC,
abstract = {Time series classification is a pivotal task in modern machine learning, with widespread applications in fields such as healthcare, finance, and cybersecurity. While deep learning methods dominate recent developments, their resource demands and privacy limitations hinder deployment on low-power and decentralized environments. To address these challenges, we introduce Fed2RC, a fully federated and gradient-free approach that integrates the efficiency of Rocket-based feature extraction with the robustness of ridge regression in a privacy-preserving setting. Fed2RC builds upon two key ideas: (i) federated selection and aggregation of high-performing random convolution kernels, and (ii) incremental and communication-efficient updates of ridge classifier parameters using closed-form solutions. Additionally, we propose a novel federated protocol for selecting the global ridge regularization parameter lambda, and show how to improve the communication efficiency by matrix factorization techniques. Extensive experiments on the UCR benchmark demonstrate that Fed2RC achieves state-of-the-art results with a fraction of the computation and communication costs. Code to reproduce the experiments can be found at: https://github.com/CasellaJr/Fed2RC.},
address = {Bologna, Italy},
author = {Bruno Casella and Samuele Fonio and Lorenzo Sciandra and Claudio Gallicchio and Marco Aldinucci and Mirko Polato and Roberto Esposito},
howpublished = {28th European Conference on Artificial Intelligence},
keywords = {icsc},
month = oct,
title = {Fed2RC: Federated Rocket Kernels and Ridge Classifier for Time Series Classification},
url = {https://datacloud.di.unito.it/index.php/s/CKpYNTrZ3KDgkc6},
year = 2025,
bdsk-url-1 = {https://datacloud.di.unito.it/index.php/s/CKpYNTrZ3KDgkc6}} Il futuro è già qui: ma noi siamo pronti?
Marco Aldinucci
@misc{25:internetfestival:future,
address = {Pisa, Italy},
author = {Aldinucci, Marco},
howpublished = {Internet Festival 2025},
month = oct,
title = {Il futuro {\`e} gi{\`a} qui: ma noi siamo pronti?},
url = {https://www.intoscana.it/it/intelligenza-artificiale-e-memoria-digitale-giornata-di-chiusura-per-internet-festival/},
year = 2025,
bdsk-url-1 = {https://www.intoscana.it/it/intelligenza-artificiale-e-memoria-digitale-giornata-di-chiusura-per-internet-festival/}} HPC-AI, dalla ricerca all'innovazione
Marco Aldinucci
ai hpc4ai icsc
@misc{25:iulm:hpcai,
address = {Milano, Italy},
author = {Marco Aldinucci},
howpublished = {Economia e Management nell'era dell'Intelligenza Artificiale, Universit\`a IULM},
keywords = {icsc, hpc4ai, ai},
month = oct,
title = {{HPC-AI}, dalla ricerca all'innovazione},
year = 2025} Parallel Computing UNITO
Marco Aldinucci
ai hpc hpc4ai icsc
@misc{25:ibm:unito,
address = {Torino, Italy},
author = {Marco Aldinucci},
howpublished = {{IBM}--Universit\`a di Torino},
keywords = {hpc4ai, icsc, hpc, ai},
month = oct,
title = {Parallel Computing \@ {UNITO}},
year = 2025} Decentralized Machine Learning and the Compute Continuum
Marco Aldinucci, Gianluca Mittone, Bruno Casella, Robert Birke
fl hpc4ai icsc
@misc{25:huawei:fl,
address = {Zurich, Switzerland},
author = {Marco Aldinucci and Gianluca Mittone and Bruno Casella and Robert Birke},
howpublished = {Huawei Snow Mountain 2025},
keywords = {icsc, hpc4ai, fl},
month = nov,
title = {Decentralized Machine Learning and the Compute Continuum},
year = 2025} 2024
2024
The impact of the advances in generative models on applications and systems
Robert Birke
ai eupilot textarossa
Generative models have achieved unprecedented quality levels across a wide range of data types. This advance often stems from the ever increasing data and compute used to train larger and larger models. One major use case of such synthetic data is in privacy-compliant data sharing. Gartner predicts that synthetic data will reduce by 2025 the need for real data by 70
@misc{24:ASF:WINTER,
abstract = {Generative models have achieved unprecedented quality levels across a wide range of data types. This advance often stems from the ever increasing data and compute used to train larger and larger models. One major use case of such synthetic data is in privacy-compliant data sharing. Gartner predicts that synthetic data will reduce by 2025 the need for real data by 70% for analytics and machine learning. We will look at generative models, with a special focus on tabular data, and the issue of democratization of large model training.},
address = {Le Pleynet, France},
annote = {keynote},
author = {Robert Birke},
howpublished = {8th GDR RSD / ASF Winter School on Distributed Systems & Networks 2024},
keywords = {eupilot, textarossa, ai},
month = jan,
note = {Keynote talk},
title = {The impact of the advances in generative models on applications and systems},
url = {https://datacloud.di.unito.it/index.php/s/QYTCMfWp4sY5qx4},
year = 2024,
bdsk-url-1 = {https://datacloud.di.unito.it/index.php/s/QYTCMfWp4sY5qx4}} RISC-V for AI
Gianluca Mittone
eupilot icsc
AI-focused RISC-V-based hardware accelerators
@misc{24:HiPEAC:riscv,
abstract = {AI-focused RISC-V-based hardware accelerators},
address = {Garching bei M{\"u}nchen, M{\"u}nchen, Germany},
author = {Mittone, Gianluca},
howpublished = {High Performance, Edge And Cloud computing Conference 2024 (HiPEAC 2024)},
keywords = {eupilot, icsc},
month = jan,
title = {RISC-V for AI},
url = {https://datacloud.di.unito.it/index.php/s/rFtxT7zryoKNGbP},
year = 2024,
bdsk-url-1 = {https://datacloud.di.unito.it/index.php/s/rFtxT7zryoKNGbP}} FLaaS: Federated Learning as a Service
Robert Birke
ai icsc
Presentation about the Innovation Grant in collaboration with IFAB
@misc{24:icsc:spoke1:ifab,
abstract = {Presentation about the Innovation Grant in collaboration with IFAB},
address = {Torino, Italy},
author = {Robert Birke},
howpublished = {ICSC - Spoke 1 meeting},
keywords = {icsc, ai},
month = feb,
title = {{FLaaS}: Federated Learning as a Service},
url = {https://datacloud.di.unito.it/index.php/s/yHXdTnC8xEqoJ6Y},
year = 2024,
bdsk-url-1 = {https://datacloud.di.unito.it/index.php/s/yHXdTnC8xEqoJ6Y}} Cross-Platform Full Waveform Inversion
Alberto Mulone
icsc streamflow
Presentation about the Innovation Grant in collaboration with ENI
@misc{24:icsc:spoke1:eni,
abstract = {Presentation about the Innovation Grant in collaboration with ENI},
address = {Torino, Italy},
author = {Mulone, Alberto},
howpublished = {ICSC - Spoke 1 meeting},
keywords = {icsc, streamflow},
month = feb,
title = {Cross-Platform Full Waveform Inversion},
url = {https://datacloud.di.unito.it/index.php/s/M3HkxA5wsBPS5ro},
year = 2024,
bdsk-url-1 = {https://datacloud.di.unito.it/index.php/s/M3HkxA5wsBPS5ro}} Benchmarking Parallelization Models through Karmarkar`s algorithm
Giulio Malenza, Marco Edoardo Santimaria
hpc icsc
Optimization problems are one of the main focus of scientific research. Their computational-intensive nature makes them prone to be parallelized with consistent improvements in performance. This paper sheds light on different parallel models for accelerating Karmarkar's Interior-point method. To do so, we assess parallelization strategies for individual operations within the aforementioned Karmarkar's algorithm using OpenMP, GPU acceleration with CUDA, and the recent Parallel Standard C++ Linear Algebra library (PSTL) executing both on GPU and CPU. Our different implementations yield interesting benchmark results that show the optimal approach for parallelizing interior point algorithms for general Linear Programming (LP) problems. In addition, we propose a more theoretical perspective of the parallelization of this algorithm, with a detailed study of our OpenMP implementation, showing the limits of optimizing the single operations
@misc{24:pdp:karmarkartalk,
abstract = {Optimization problems are one of the main focus of scientific research. Their computational-intensive nature makes them prone to be parallelized with consistent improvements in performance. This paper sheds light on different parallel models for accelerating Karmarkar's Interior-point method. To do so, we assess parallelization strategies for individual operations within the aforementioned Karmarkar's algorithm using OpenMP, GPU acceleration with CUDA, and the recent Parallel Standard C++ Linear Algebra library (PSTL) executing both on GPU and CPU. Our different implementations yield interesting benchmark results that show the optimal approach for parallelizing interior point algorithms for general Linear Programming (LP) problems. In addition, we propose a more theoretical perspective of the parallelization of this algorithm, with a detailed study of our OpenMP implementation, showing the limits of optimizing the single operations},
address = {Dublin, Irelans},
author = {Malenza, Giulio and Santimaria, Marco Edoardo},
keywords = {icsc, hpc},
month = mar,
title = {Benchmarking Parallelization Models through Karmarkar`s algorithm},
url = {https://datacloud.di.unito.it/index.php/s/JjKcAJpYS7ctX9r},
year = 2024,
bdsk-url-1 = {https://datacloud.di.unito.it/index.php/s/JjKcAJpYS7ctX9r}} Parallel Computing at the University of Torino
Marco Aldinucci
ai fl hpc hpc4ai icsc streamflow
@misc{24:alpha:unito,
author = {Marco Aldinucci},
howpublished = {University of Torino},
keywords = {hpc4ai, icsc, hpc, ai, fl, streamflow},
month = mar,
title = {Parallel Computing at the University of Torino},
year = 2024} Secure Generic Remote Workflow Execution with TEEs
Lorenzo Brescia
confidential icsc
In scientific environments, the frequent need to process substantial volumes of data poses a common challenge. Individuals tasked with executing these computations frequently encounter a deficit in local computational resources, leading them to opt for the facilities of a Cloud Service Provider (CSP) for data processing. However, the data subjected to these calculations may be subject to confidentiality constraints. This paper introduces a proof-of-concept framework that leverages Gramine LibOS and Intel SGX, enabling the protection of generic remote workflow computations through SGX enclaves as Trusted Execution Environments (TEEs). The framework entails the delineation of user and CSP behavior and has been implemented using Bash scripts. Furthermore, an infrastructure has been designed for the Data Center Attestation Primitives (DCAP) remote attestation mechanism, wherein the user gains trust in the proper instantiation of the enclave within the CSP. To assess the framework efficacy, it has been tested on two distinct workflows, one trivial and the other involving real-world bioinformatics applications for processing DNA data. The performance study revealed that the framework incurred an acceptable overhead, ranging from a factor of x1.4 to x1.8 compared to unsafe execution practice.
@misc{23:brescia:wide:talk,
abstract = {In scientific environments, the frequent need to process substantial volumes of data poses a common challenge. Individuals tasked with executing these computations frequently encounter a deficit in local computational resources, leading them to opt for the facilities of a Cloud Service Provider (CSP) for data processing. However, the data subjected to these calculations may be subject to confidentiality constraints. This paper introduces a proof-of-concept framework that leverages Gramine LibOS and Intel SGX, enabling the protection of generic remote workflow computations through SGX enclaves as Trusted Execution Environments (TEEs). The framework entails the delineation of user and CSP behavior and has been implemented using Bash scripts. Furthermore, an infrastructure has been designed for the Data Center Attestation Primitives (DCAP) remote attestation mechanism, wherein the user gains trust in the proper instantiation of the enclave within the CSP. To assess the framework efficacy, it has been tested on two distinct workflows, one trivial and the other involving real-world bioinformatics applications for processing DNA data. The performance study revealed that the framework incurred an acceptable overhead, ranging from a factor of x1.4 to x1.8 compared to unsafe execution practice.},
address = {Athens, Greece},
author = {Brescia, Lorenzo},
howpublished = {{P}roceedings of the 2nd {W}orkshop on {W}orkflows in {D}istributed {E}nvironments},
keywords = {icsc, confidential},
month = apr,
title = {{S}ecure {G}eneric {R}emote {W}orkflow {E}xecution with {TEE}s},
url = {https://datacloud.di.unito.it/index.php/s/Prxq6EWGbcN8sWx},
year = 2024,
bdsk-url-1 = {https://datacloud.di.unito.it/index.php/s/Prxq6EWGbcN8sWx}} Workflows for future High-Performance Computing
Alberto Mulone
icsc
@misc{24:amulone:comete,
address = {Torino, Italy},
annote = {pitch talk},
author = {Mulone, Alberto},
howpublished = {COMETE PhD Workshop},
keywords = {icsc},
month = apr,
title = {Workflows for future High-Performance Computing},
url = {https://datacloud.di.unito.it/index.php/s/ZGG8fLMp5B7qRHS},
year = 2024,
bdsk-url-1 = {https://datacloud.di.unito.it/index.php/s/ZGG8fLMp5B7qRHS}} CWL in the HPC Ecosystem
Iacopo Colonnelli
across eupex icsc space streamflow
@misc{24:icolonne:cwl4hpccern,
address = {CERN, Meyrin, Switzerland},
author = {Colonnelli, Iacopo},
howpublished = {Workshop on workflow languages for {HEP} analysis},
keywords = {across, eupex, space, icsc, streamflow},
month = apr,
title = {CWL in the HPC Ecosystem},
url = {https://datacloud.di.unito.it/index.php/s/PRmqdwWHt6P2PH7},
year = 2024,
bdsk-url-1 = {https://datacloud.di.unito.it/index.php/s/PRmqdwWHt6P2PH7}} Cloud Computing Hands-on the OpenStack Cloud
Marco Aldinucci
hpc4ai openscience
@misc{24:unito:openstack,
address = {Torino, Italy},
author = {Marco Aldinucci},
howpublished = {Data Steward per Open Science, Universit\`a di Torino},
keywords = {openscience, hpc4ai},
month = apr,
title = {Cloud Computing Hands-on the OpenStack Cloud},
year = 2024} Dynamic hybrid workflows for Deep Learning on HPC infrastructure
Iacopo Colonnelli
icsc jupyter-workflow streamflow
Hybrid workflow abstractions allow users to quickly design and orchestrate cross-facility workloads, decoupling tasks from environment-specific technical details to reduce complexity and increase reusability. Plus, workflow descriptions help ensure the reproducibility of scientific experiments through prospective and retrospective provenance collection. This module has been designed to provide a hands-on exploration of scientific workflows from various angles, from the initial design phase to their orchestration at extreme scales. We will use the practical example of the CommonWorkflow Language (CWL) open standard to demonstrate how workflows can be written, and the StreamFlow workflow system to execute them seamlessly on the CINECA HPC facility. We will also delve into the integration between scientific workflows and Jupyter Notebooks, which aims to give data scientists a familiar interface to scientific workflows. In this module, students will gain a comprehensive understanding of scientific workflows. They will learn how to use these workflows to model and orchestrate Machine Learning and Deep Learning pipelines. Additionally, they will explore how modern workflow management systems can efficiently scale data-oriented workloads from a researcher's laptop to an entire HPC facility.
@misc{24:icolonne:ictp,
abstract = {Hybrid workflow abstractions allow users to quickly design and orchestrate cross-facility workloads, decoupling tasks from environment-specific technical details to reduce complexity and increase reusability. Plus, workflow descriptions help ensure the reproducibility of scientific experiments through prospective and retrospective provenance collection. This module has been designed to provide a hands-on exploration of scientific workflows from various angles, from the initial design phase to their orchestration at extreme scales. We will use the practical example of the CommonWorkflow Language (CWL) open standard to demonstrate how workflows can be written, and the StreamFlow workflow system to execute them seamlessly on the CINECA HPC facility. We will also delve into the integration between scientific workflows and Jupyter Notebooks, which aims to give data scientists a familiar interface to scientific workflows. In this module, students will gain a comprehensive understanding of scientific workflows. They will learn how to use these workflows to model and orchestrate Machine Learning and Deep Learning pipelines. Additionally, they will explore how modern workflow management systems can efficiently scale data-oriented workloads from a researcher's laptop to an entire HPC facility.},
address = {Trieste, Italy},
author = {Iacopo Colonnelli},
keywords = {icsc, streamflow, jupyter-workflow},
month = may,
title = {Dynamic hybrid workflows for Deep Learning on HPC infrastructure},
url = {https://datacloud.di.unito.it/index.php/s/EaFHJEKNbW5oXeq},
year = 2024,
bdsk-url-1 = {https://datacloud.di.unito.it/index.php/s/EaFHJEKNbW5oXeq}} Into to Federated Learning
Gianluca Mittone
fl icsc
Machine Learning (ML) is the branch of Artificial Intelligence focused on developing algorithms capable of adapting and improving their predictive or generative performance by feeding on data. Adapting or improving the system's behaviour based on the provided data is called learning since it is similar to the human learning process in many aspects. The same ML algorithm, usually referred to as a model, trained on different data will thus expose different capabilities and can, therefore, solve different tasks. FL is a relatively recent distributed ML methodology aiming to bridge the gap between the need to train ever bigger ML models on ever larger datasets and the individual and companies' will to protect and not share their private data. From another point of view, FL is also a way to distribute the training of an ML model even more than before. However, it should be considered that the learning performance of FL is usually lower than that of traditional centralised learning. This course will start from Kairouz ad McMahan's definition of FL: ”Federated learning is a machine learning setting where multiple entities (clients) collaborate in solving a machine learning problem, under the coordination of a central server or service provider. Each client's raw data is stored locally and not exchanged or transferred; instead, focused updates intended for immediate aggregation are used to achieve the learning objective.” From this starting point, the most significant aspects of FL will be exposed and discussed. This tutorial will particularly explore FL from both the learning and computational [5] performance perspectives, investigating its pros and cons in a distributed ML setting. Since FL natively targets data privacy, some insights on how the FL process can be attacked and protected will also be discussed from a high-level perspective. Finally, a hands-on session will guide the participants in building a basic FL system, providing a better understanding of the major implementational difficulties of such a technique.
@misc{24:mittone:ictp,
abstract = {Machine Learning (ML) is the branch of Artificial Intelligence focused on developing algorithms capable of adapting and improving their predictive or generative performance by feeding on data. Adapting or improving the system's behaviour based on the provided data is called learning since it is similar to the human learning process in many aspects. The same ML algorithm, usually referred to as a model, trained on different data will thus expose different capabilities and can, therefore, solve different tasks. FL is a relatively recent distributed ML methodology aiming to bridge the gap between the need to train ever bigger ML models on ever larger datasets and the individual and companies' will to protect and not share their private data. From another point of view, FL is also a way to distribute the training of an ML model even more than before. However, it should be considered that the learning performance of FL is usually lower than that of traditional centralised learning. This course will start from Kairouz ad McMahan's definition of FL: ''Federated learning is a machine learning setting where multiple entities (clients) collaborate in solving a machine learning problem, under the coordination of a central server or service provider. Each client's raw data is stored locally and not exchanged or transferred; instead, focused updates intended for immediate aggregation are used to achieve the learning objective.'' From this starting point, the most significant aspects of FL will be exposed and discussed. This tutorial will particularly explore FL from both the learning and computational [5] performance perspectives, investigating its pros and cons in a distributed ML setting. Since FL natively targets data privacy, some insights on how the FL process can be attacked and protected will also be discussed from a high-level perspective. Finally, a hands-on session will guide the participants in building a basic FL system, providing a better understanding of the major implementational difficulties of such a technique.},
address = {Trieste, Italy},
author = {Gianluca Mittone},
keywords = {icsc, fl},
month = may,
title = {Into to Federated Learning},
url = {https://datacloud.di.unito.it/index.php/s/nSwwmedjqe2jbWJ},
year = 2024,
bdsk-url-1 = {https://datacloud.di.unito.it/index.php/s/nSwwmedjqe2jbWJ}} CWL Working Groups
Iacopo Colonnelli
icsc
This presentation introduces the new CWL Working Groups initiative, describing what a Working Group actually is, which Working Groups already exist in the CWL community, and how anybody can create a new officially recognized Working Group. Then, the presentation will explore the CWL4HPC Working Group, using it as an example of how a CWL Working Group can actually work.
@misc{24:icolonne:cwlcon2024,
abstract = {This presentation introduces the new CWL Working Groups initiative, describing what a Working Group actually is, which Working Groups already exist in the CWL community, and how anybody can create a new officially recognized Working Group. Then, the presentation will explore the CWL4HPC Working Group, using it as an example of how a CWL Working Group can actually work.},
address = {Amsterdam, Netherlands},
author = {Iacopo Colonnelli},
howpublished = {2024 CWL Conference},
keywords = {icsc},
month = may,
title = {{CWL} Working Groups},
url = {https://datacloud.di.unito.it/index.php/s/zZDKdL8deLd4jSi},
year = 2024,
bdsk-url-1 = {https://datacloud.di.unito.it/index.php/s/zZDKdL8deLd4jSi}} Towards Secure WMS with TEEs
Lorenzo Brescia
confidential icsc
@misc{24:brescia:hpc:school:talk,
address = {Trento, Italy},
author = {Brescia, Lorenzo},
howpublished = {{Mentoring session: HPC Summer School, university of Trento}},
keywords = {icsc, confidential},
month = jun,
title = {{Towards Secure WMS with TEEs}},
url = {https://datacloud.di.unito.it/index.php/s/eAxEgqiTsGSRQz4},
year = 2024,
bdsk-url-1 = {https://datacloud.di.unito.it/index.php/s/eAxEgqiTsGSRQz4}} Cross-Facility Federated Learning - Part II
Iacopo Colonnelli, Robert Birke, Giulio Malenza, Gianluca Mittone, Alberto Mulone, Marco Aldinucci
eupex icsc space
@misc{24:ic:elise:xffl,
address = {Helsinki, Finland},
author = {Iacopo Colonnelli and Robert Birke and Giulio Malenza and Gianluca Mittone and Alberto Mulone and Marco Aldinucci},
keywords = {icsc,space,eupex},
month = jun,
note = {Invited talk},
title = {Cross-Facility Federated Learning - Part II},
url = {https://datacloud.di.unito.it/index.php/s/7HonBpcWPxotXLX},
year = 2024,
bdsk-url-1 = {https://datacloud.di.unito.it/index.php/s/7HonBpcWPxotXLX}} From HPC4AI to Software & Integration living lab to innovation
Marco Aldinucci
ai hpc icsc
The talk presents the motivation and the activity of the "Software and Integration" lab at UNITO.
@misc{24:ma:hpcai:talk,
abstract = {The talk presents the motivation and the activity of the "Software and Integration" lab at UNITO.},
address = {Torino, Italy},
author = {Marco Aldinucci},
howpublished = {HPC as an enabling platform for AI event},
keywords = {icsc,ai,hpc},
month = jun,
title = {From HPC4AI to Software & Integration living lab to innovation},
url = {https://datacloud.di.unito.it/index.php/s/3SS3Xa9XorN6D9o},
year = 2024,
bdsk-url-1 = {https://datacloud.di.unito.it/index.php/s/3SS3Xa9XorN6D9o}} Software & Integration lab of FutureHPC spoke
Marco Aldinucci
ai hpc icsc
The presentation describes the main the activity of the "Software and Integration" lab at UNITO across its main flagship codes.
@misc{24:ma:swi:lab,
abstract = {The presentation describes the main the activity of the "Software and Integration" lab at UNITO across its main flagship codes.},
address = {Torino, Italy},
author = {Marco Aldinucci},
howpublished = {HPC as an enabling platform for AI event},
keywords = {icsc,ai,hpc},
month = jun,
title = {Software & Integration lab of FutureHPC spoke},
url = {https://datacloud.di.unito.it/index.php/s/WDjyXCGyYFJDQSd},
year = 2024,
bdsk-url-1 = {https://datacloud.di.unito.it/index.php/s/WDjyXCGyYFJDQSd}} Preliminary analysis of model parallelism applications on a 64-core RV64 Server CPU
Giulio Malenza
eupilot icsc
Massive Data Parallel workloads, driven by inference on large ML models, are pushing hardware vendors to develop efficient and cost-effective multi-core server CPUs. The RISC-V architecture plays a prominent role due to its open, extensible and energy-friendly ISA. Despite significant progress in recent years, finding efficient methods to run parallel applications on new architectures to harness their maximum performance fully remains a challenge. In this study, we benchmark the inference of machine learning models on the SOPHON SG2042 SoC, the first server-grade CPU based on the RV64 ISA, composed of 64 cores arranged in a grid of 16 groups of 4 cores. Specifically, we aim to enhance performance via better cache hit ratios stemming from model parallelism to split and assign parts of the model to specific (groups of) cores using a pipeline execution. We orchestrate execution using FastFlow, a low-level programming framework designed for multithreaded streaming applications. By comparing the results against the standard multi-core inference and analyzing the effects of different submodel-to-core mapping strategies, we aim to provide a comprehensive understanding of how the model parallel approach can maximize efficiency and utilization of hardware resources.
@misc{24:gmalenza:hlpp:MPRISC-v,
abstract = {Massive Data Parallel workloads, driven by inference on large ML models, are pushing hardware vendors to develop efficient and cost-effective multi-core server CPUs. The RISC-V architecture plays a prominent role due to its open, extensible and energy-friendly ISA. Despite significant progress in recent years, finding efficient methods to run parallel applications on new architectures to harness their maximum performance fully remains a challenge. In this study, we benchmark the inference of machine learning models on the SOPHON SG2042 SoC, the first server-grade CPU based on the RV64 ISA, composed of 64 cores arranged in a grid of 16 groups of 4 cores. Specifically, we aim to enhance performance via better cache hit ratios stemming from model parallelism to split and assign parts of the model to specific (groups of) cores using a pipeline execution. We orchestrate execution using FastFlow, a low-level programming framework designed for multithreaded streaming applications. By comparing the results against the standard multi-core inference and analyzing the effects of different submodel-to-core mapping strategies, we aim to provide a comprehensive understanding of how the model parallel approach can maximize efficiency and utilization of hardware resources.},
address = {Pisa, Italy},
author = {Giulio Malenza},
keywords = {icsc,eupilot},
month = jul,
title = {Preliminary analysis of model parallelism applications on a 64-core RV64 Server CPU},
url = {https://datacloud.di.unito.it/index.php/s/JrWwKALeaFEJSQo},
year = 2024,
bdsk-url-1 = {https://datacloud.di.unito.it/index.php/s/JrWwKALeaFEJSQo}} CAPIO-CL: Cross Application Programmable IO - Coordination Language
Marco Edoardo Santimaria
across admire capio eupex icsc
The performance bottleneck in file-based workflows remains a pressing issue in the realm of I/O-based workflows. To address this challenge, a novel annotation language has been developed. CAPIO-CL is positioned as an innovative I/O coordination language, enabling users to annotate data dependencies within file-based workflows with synchronization semantics pertinent to the involved files and directories. Through the information provided by the language, optimization opportunities arise in streaming and preemptive data movement. This paper serves to illustrate the semantics and syntax enabling CAPIO-CL to enhance the performance of in situ workflows without necessitating the rewriting or modification of the original workflow application steps. Finally, an analysis of CAPIO-CL is provided, taking into consideration both language expressiveness and application performance enhancement.
@misc{24:santimaria:hlpp:capiocl,
abstract = {The performance bottleneck in file-based workflows remains a pressing issue in the realm of I/O-based workflows. To address this challenge, a novel annotation language has been developed. CAPIO-CL is positioned as an innovative I/O coordination language, enabling users to annotate data dependencies within file-based workflows with synchronization semantics pertinent to the involved files and directories. Through the information provided by the language, optimization opportunities arise in streaming and preemptive data movement. This paper serves to illustrate the semantics and syntax enabling CAPIO-CL to enhance the performance of in situ workflows without necessitating the rewriting or modification of the original workflow application steps. Finally, an analysis of CAPIO-CL is provided, taking into consideration both language expressiveness and application performance enhancement.},
address = {Pisa, Italy},
author = {Marco Edoardo Santimaria},
date-modified = {2026-09-19 19:50:32 +0200},
keywords = {icsc,capio,across,eupex,admire},
month = jul,
title = {{CAPIO-CL}: Cross Application Programmable IO - Coordination Language},
url = {https://datacloud.di.unito.it/index.php/s/zsKY3PWzX5NFCiX},
year = 2024,
bdsk-url-1 = {https://datacloud.di.unito.it/index.php/s/zsKY3PWzX5NFCiX}} A Fault Tolerance mechanism for Hybrid Scientific Workflows
Alberto Mulone, Doriana Medić, Marco Aldinucci
eupex icsc streamflow
In large distributed systems, failures are a daily event occurring frequently, especially with growing numbers of computation tasks and locations on which they are deployed. The advantage of representing an application as a workflow is possibility to utilize the Workflow Management Systems which are reliable systems guaranteeing the correct execution of the application and providing the features such as portability, scalability, and fault tolerance. Over recent years, the emergence of hybrid workflows has posed new and intriguing challenges by increasing the possibility of distributing computations involving heterogeneous and independent environments. As a consequence, the number of possible points of failure in the execution augmented, creating different important challenges interesting to study.
@misc{24:madrid:hipes:talk,
abstract = {In large distributed systems, failures are a daily event occurring frequently, especially with growing numbers of computation tasks and locations on which they are deployed. The advantage of representing an application as a workflow is possibility to utilize the Workflow Management Systems which are reliable systems guaranteeing the correct execution of the application and providing the features such as portability, scalability, and fault tolerance. Over recent years, the emergence of hybrid workflows has posed new and intriguing challenges by increasing the possibility of distributing computations involving heterogeneous and independent environments. As a consequence, the number of possible points of failure in the execution augmented, creating different important challenges interesting to study.},
address = {Madrid, Spain},
author = {Alberto Mulone and Doriana Medi\'{c} and Marco Aldinucci},
howpublished = {1st workshop about {H}igh-{P}erformance e-{S}cience (HiPES)},
keywords = {streamflow, icsc, eupex},
month = aug,
title = {A Fault Tolerance mechanism for Hybrid Scientific Workflows},
url = {https://datacloud.di.unito.it/index.php/s/9Ddj6fGgmDbLDXj},
year = 2024,
bdsk-url-1 = {https://datacloud.di.unito.it/index.php/s/9Ddj6fGgmDbLDXj}} Scientific Workflows in the Continuum Era
Iacopo Colonnelli
icsc
Thanks to their generality, workflow models represent a powerful abstraction for designing complex applications and executing them on large-scale distributed architectures. However, several additional challenges appear when transitioning from cloud/HPC environments to the entire compute continuum. Continuum execution environments are fully distributed and modular, and modules can be heterogeneous and independent of each other. In addition, continuum workflows often rely on multiple intercommunicating agents that form complex micro-services architectures. Different agents deal with different communication and parallelization paradigms: network-based stream processing at the edge and file-based batch processing on HPC facilities. Finally, support for efficient interactive workflows in the continuum remains an open research problem. This talk explores these challenges and provides insights on how to deal with them. A ready-to-use software library accompanies each proposed solution to facilitate the reproducibility and reusability of the presented concepts.
@misc{24:icolonne:wscc,
abstract = {Thanks to their generality, workflow models represent a powerful abstraction for designing complex applications and executing them on large-scale distributed architectures. However, several additional challenges appear when transitioning from cloud/HPC environments to the entire compute continuum. Continuum execution environments are fully distributed and modular, and modules can be heterogeneous and independent of each other. In addition, continuum workflows often rely on multiple intercommunicating agents that form complex micro-services architectures. Different agents deal with different communication and parallelization paradigms: network-based stream processing at the edge and file-based batch processing on HPC facilities. Finally, support for efficient interactive workflows in the continuum remains an open research problem. This talk explores these challenges and provides insights on how to deal with them. A ready-to-use software library accompanies each proposed solution to facilitate the reproducibility and reusability of the presented concepts.},
address = {Madrid, Spain},
author = {Iacopo Colonnelli},
keywords = {icsc},
month = aug,
note = {Keynote Talk},
title = {Scientific Workflows in the Continuum Era},
url = {https://datacloud.di.unito.it/index.php/s/PkqYA3p38XLKgrt},
year = 2024,
bdsk-url-1 = {https://datacloud.di.unito.it/index.php/s/PkqYA3p38XLKgrt}} Assessing Large Language Models Inference Performance on a 64-core RISC-V CPU with Silicon-Enabled Vectors
Adriano Garcia Marques
confidential icsc
@misc{24:garcia:bigHPC:talk,
address = {Pisa, Italy},
author = {Marques, Adriano Garcia},
howpublished = {{Proceedings of BigHPC2024: Special Track on Big Data and High-Performance Computing, co-located with the 3\textsuperscript{rd} Italian Conference on Big Data and Data Science, ITADATA2024}},
keywords = {icsc, confidential},
month = sep,
title = {{Assessing Large Language Models Inference Performance on a 64-core RISC-V CPU with Silicon-Enabled Vectors}},
url = {https://datacloud.di.unito.it/index.php/s/eN6Z62RQr2QsRYa},
year = 2024,
bdsk-url-1 = {https://datacloud.di.unito.it/index.php/s/eN6Z62RQr2QsRYa}} Performance Analysis on DNA Alignment Workload with Intel SGX Multithreading
Lorenzo Brescia
confidential icsc
@misc{24:brescia:bigHPC:talk,
address = {Pisa, Italy},
author = {Brescia, Lorenzo},
howpublished = {{Proceedings of BigHPC2024: Special Track on Big Data and High-Performance Computing, co-located with the 3\textsuperscript{rd} Italian Conference on Big Data and Data Science, ITADATA2024}},
keywords = {icsc, confidential},
month = sep,
title = {{Performance Analysis on DNA Alignment Workload with Intel SGX Multithreading}},
url = {https://datacloud.di.unito.it/index.php/s/wqgYiKpHBw5zbSa},
year = 2024,
bdsk-url-1 = {https://datacloud.di.unito.it/index.php/s/wqgYiKpHBw5zbSa}} Scientific Workflows in the Heterogeneous Computing Era
Iacopo Colonnelli
icsc
@misc{24:icolonne:ICSC,
address = {Roma, Italy},
author = {Iacopo Colonnelli},
keywords = {icsc},
month = sep,
title = {Scientific Workflows in the Heterogeneous Computing Era},
url = {https://datacloud.di.unito.it/index.php/s/CyxiWsDbdg6rbpQ},
year = 2024,
bdsk-url-1 = {https://datacloud.di.unito.it/index.php/s/CyxiWsDbdg6rbpQ}} Exploiting C++ Parallel Algorithms through FastFlow
Giulio Malenza
icsc
High-performance computing and artificial intelligent simulations necessitate the rapid processing of large quantities of data. To handle such data volumes efficiently, leveraging the parallelism inherent in algorithms is crucial. Consequently, parallel programming frameworks have been developed to fully exploit modern parallel architectures. Among these, C++ PSTL stands out for its user-friendliness, portability, and high performance. In this study, we introduce a back-end for the PSTL implemented using the FastFlow parallel programming framework. We will evaluate correctness and performance of the back-end comparing results with other coming from traditional vendor-dependent back-ends like TBB and nvc++. Performance metrics are derived from running the LULESH application on both RISC-V and ARM architectures. Our results indicate that all three back-ends deliver comparable performance.
@misc{24:gmalenza:BigHPC2024,
abstract = {High-performance computing and artificial intelligent simulations necessitate the rapid processing of large quantities of data. To handle such data volumes efficiently, leveraging the parallelism inherent in algorithms is crucial. Consequently, parallel programming frameworks have been developed to fully exploit modern parallel architectures. Among these, C++ PSTL stands out for its user-friendliness, portability, and high performance. In this study, we introduce a back-end for the PSTL implemented using the FastFlow parallel programming framework. We will evaluate correctness and performance of the back-end comparing results with other coming from traditional vendor-dependent back-ends like TBB and nvc++. Performance metrics are derived from running the LULESH application on both RISC-V and ARM architectures. Our results indicate that all three back-ends deliver comparable performance.},
address = {Pisa, Italy},
author = {Malenza, Giulio},
keywords = {icsc},
month = sep,
title = {Exploiting C++ Parallel Algorithms through FastFlow},
url = {https://datacloud.di.unito.it/index.php/s/GcpQ8cz9BRyM85B},
year = 2024,
bdsk-url-1 = {https://datacloud.di.unito.it/index.php/s/GcpQ8cz9BRyM85B}} Exploring energy consumption of AI frameworks on a 64-core RV64 Server CPU
Giulio Malenza
ai DYMAN icsc
In today's era of rapid technological advancement, artificial intelligence (AI) applications require large-scale, high-performance, and data-intensive computations, leading to significant energy demands. Addressing this challenge necessitates a combined approach involving both hardware and software innovations. Hardware manufacturers are developing new, efficient, and specialized solutions, with the RISC-V architecture emerging as a prominent player due to its open, extensible, and energy-efficient instruction set architecture (ISA). Simultaneously, software developers are creating new algorithms and frameworks, yet their energy efficiency often remains unclear. In this study, we conduct a comprehensive benchmark analysis of machine learning (ML) applications on the 64-core SOPHON SG2042 RISC-V architecture. Specifically, we examine the energy consumption of deep learning inference models across various AI frameworks. By comparing the performance of different frameworks, we aim to provide a detailed understanding of how these frameworks can optimize energy consumption on this architecture.
@misc{24:gmalenza:scihpcexa,
abstract = {In today's era of rapid technological advancement, artificial intelligence (AI) applications require large-scale, high-performance, and data-intensive computations, leading to significant energy demands. Addressing this challenge necessitates a combined approach involving both hardware and software innovations. Hardware manufacturers are developing new, efficient, and specialized solutions, with the RISC-V architecture emerging as a prominent player due to its open, extensible, and energy-efficient instruction set architecture (ISA). Simultaneously, software developers are creating new algorithms and frameworks, yet their energy efficiency often remains unclear. In this study, we conduct a comprehensive benchmark analysis of machine learning (ML) applications on the 64-core SOPHON SG2042 RISC-V architecture. Specifically, we examine the energy consumption of deep learning inference models across various AI frameworks. By comparing the performance of different frameworks, we aim to provide a detailed understanding of how these frameworks can optimize energy consumption on this architecture.},
address = {Pisa, Italy},
author = {Malenza, Giulio},
keywords = {icsc, DYMAN, ai},
month = sep,
title = {Exploring energy consumption of AI frameworks on a 64-core RV64 Server CPU},
url = {https://datacloud.di.unito.it/index.php/s/5aTdyzNB6n9CREq},
year = 2024,
bdsk-url-1 = {https://datacloud.di.unito.it/index.php/s/5aTdyzNB6n9CREq}} Releasing the CAPIO middleware from MPI derived constraints
Marco Edoardo Santimaria, Iacopo Colonnelli, Marco Aldinucci
across admire capio capiocl eupex icsc
CAPIO is a middleware that transparently injects streaming capabilities into file-based workflows. However, its implementation is limited to HPC environments based on the MPI framework, significantly limiting its applications. This paper will illustrate a proposed architecture and some preliminary results aimed at investigating the usage of a distributed files system as a communication media for the CAPIO middleware, with the ultimate goal of supporting both CLOUD-based and HPC-based workflows.
@misc{24:santimaria:bighpc,
abstract = {CAPIO is a middleware that transparently injects streaming capabilities into file-based workflows. However, its implementation is limited to HPC environments based on the MPI framework, significantly limiting its applications. This paper will illustrate a proposed architecture and some preliminary results aimed at investigating the usage of a distributed files system as a communication media for the CAPIO middleware, with the ultimate goal of supporting both CLOUD-based and HPC-based workflows.},
address = {Pisa, Italy},
author = {Santimaria, Marco Edoardo and Colonnelli, Iacopo and Aldinucci, Marco},
keywords = {icsc, capio, capiocl, across, admire, eupex},
month = sep,
title = {Releasing the CAPIO middleware from MPI derived constraints},
url = {https://datacloud.di.unito.it/index.php/s/zrJGD4i36fWdp5g},
year = 2024,
bdsk-url-1 = {https://datacloud.di.unito.it/index.php/s/zrJGD4i36fWdp5g}} CAPIO: Cross Application Programamble IO
Marco Edoardo Santimaria, Iacopo Colonnelli, Massimo Torquati, Marco Aldinucci
across admire capio capiocl eupex icsc
With the increasing amount of digital data available for analysis and simulation, the class of I/O-intensive HPC workflows is fated to expand, further exacerbating quickly the performance gap between computing, memory, and storage technologies. CAPIO (Cross-Application Programmable I/O), is a middleware capable of injecting I/O streaming capabilities into file-based workflows, improving the computation-I/O overlap without the need to change the application code. In this presentation, we will introduce the CAPIO-CL language with its semantics, as well as the implementation of the CAPIO-CL language through the CAPIO middleware. We will also provide some case studies of how CAPIO has been employed to improve workflow execution time as well as some future directions.
@misc{24:santimaria:itadata:shpcpee,
abstract = {With the increasing amount of digital data available for analysis and simulation, the class of I/O-intensive HPC workflows is fated to expand, further exacerbating quickly the performance gap between computing, memory, and storage technologies. CAPIO (Cross-Application Programmable I/O), is a middleware capable of injecting I/O streaming capabilities into file-based workflows, improving the computation-I/O overlap without the need to change the application code. In this presentation, we will introduce the CAPIO-CL language with its semantics, as well as the implementation of the CAPIO-CL language through the CAPIO middleware. We will also provide some case studies of how CAPIO has been employed to improve workflow execution time as well as some future directions.},
address = {Pisa, Italy},
author = {Santimaria, Marco Edoardo and Colonnelli, Iacopo and Torquati, Massimo and Aldinucci, Marco},
keywords = {icsc, capio, capiocl, across, admire, eupex},
month = sep,
title = {CAPIO: Cross Application Programamble IO},
url = {https://datacloud.di.unito.it/index.php/s/rg6LWwrZXi6tTXm},
year = 2024,
bdsk-url-1 = {https://datacloud.di.unito.it/index.php/s/rg6LWwrZXi6tTXm}} Benchmarking HPC Performance for State-of-the-Art AI Workloads
Gianluca Mittone
ai eupilot icsc
Benchmarking the performance of modern High-Performance Computing (HPC) infrastructure on Artificial Intelligence (AI) workloads is a hot topic in the supercomputing community. While research communities and big-tech companies actively invest in larger, more powerful data centres to support AI research, the standard computational performance benchmarking tools (e.g., LINPACK) are increasingly becoming outdated since they are not specifically tailored for AI workloads. Some tools, such as MLPerf, are trying to bridge this gap, but the HPC community still has not adopted them as standards. Since this trend became particularly evident with the advent of Large Language Models (LLMs), this work will delve into LLM training at scale as a way to benchmark Top500 HPC infrastructures on current AI workloads. The scalability performances of a major LLM model (i.e., Meta's LLaMA) on different HPCs (Leonardo, LUMI, MeluXina, Karolina) are exposed and discussed along with their Top500 positioning. However, it should be noted that state-of-the-art LLM models are not trained on thousands of computing nodes but on hundreds. This choice is due to multiple factors, such as the influence of the training scaling on the model's convergence and the instability of large-scale deployments due to hardware/software failure. A benchmarking approach based on the next-generation LLM training approach is proposed to bypass all these issues. State-of-the-art LLMs are not monolithic structures but Mixture-of-Experts (MoE) models; this design implies innovative frontiers for the distributed training of such models due to the experts' training being potentially more parallelisable than a single monolithic model. We thus propose to create an AI-oriented HPC benchmark suite based on the parallel training of MoE models to measure the throughput performance of HPC systems on state-of-the-art AI workloads.
@misc{24:mittone:itadata:shpcpee,
abstract = {Benchmarking the performance of modern High-Performance Computing (HPC) infrastructure on Artificial Intelligence (AI) workloads is a hot topic in the supercomputing community. While research communities and big-tech companies actively invest in larger, more powerful data centres to support AI research, the standard computational performance benchmarking tools (e.g., LINPACK) are increasingly becoming outdated since they are not specifically tailored for AI workloads. Some tools, such as MLPerf, are trying to bridge this gap, but the HPC community still has not adopted them as standards. Since this trend became particularly evident with the advent of Large Language Models (LLMs), this work will delve into LLM training at scale as a way to benchmark Top500 HPC infrastructures on current AI workloads. The scalability performances of a major LLM model (i.e., Meta's LLaMA) on different HPCs (Leonardo, LUMI, MeluXina, Karolina) are exposed and discussed along with their Top500 positioning. However, it should be noted that state-of-the-art LLM models are not trained on thousands of computing nodes but on hundreds. This choice is due to multiple factors, such as the influence of the training scaling on the model's convergence and the instability of large-scale deployments due to hardware/software failure. A benchmarking approach based on the next-generation LLM training approach is proposed to bypass all these issues. State-of-the-art LLMs are not monolithic structures but Mixture-of-Experts (MoE) models; this design implies innovative frontiers for the distributed training of such models due to the experts' training being potentially more parallelisable than a single monolithic model. We thus propose to create an AI-oriented HPC benchmark suite based on the parallel training of MoE models to measure the throughput performance of HPC systems on state-of-the-art AI workloads.},
address = {Pisa, Italy},
author = {Mittone, Gianluca},
keywords = {icsc, eupilot, ai},
month = sep,
title = {Benchmarking HPC Performance for State-of-the-Art AI Workloads},
url = {https://datacloud.di.unito.it/index.php/s/5Ep3W7cPW5baZfr},
year = 2024,
bdsk-url-1 = {https://datacloud.di.unito.it/index.php/s/5Ep3W7cPW5baZfr}} Federated Adaboost for Survival Analysis
Oussama Harrak, Bruno Casella, Samuele Fonio, Piero Fariselli, Gianluca Mittone, Cesare Rollo, Tiziana Sanavia, Marco Aldinucci
ai epi fl icsc
This work proposes FedSurvBoost, a federated learning pipeline for survival analysis based on the AdaBoost.F algorithm, which iteratively aggregates the best local weak hypotheses. Our method extends AdaBoost.F by removing the dependence on the number of classes coefficient from the computation of the weights of the best model. This makes it suitable for regression tasks, such as survival analysis. We show the effectiveness of our approach by comparing it with state-of-the-art methods, specifically developed for survival analysis problems, on two common survival datasets.
@misc{23:ecmlpkdd:fedsurvboost,
abstract = {This work proposes FedSurvBoost, a federated learning pipeline for survival analysis based on the AdaBoost.F algorithm, which iteratively aggregates the best local weak hypotheses. Our method extends AdaBoost.F by removing the dependence on the number of classes coefficient from the computation of the weights of the best model. This makes it suitable for regression tasks, such as survival analysis. We show the effectiveness of our approach by comparing it with state-of-the-art methods, specifically developed for survival analysis problems, on two common survival datasets.},
address = {Vilnius, Lithuania},
author = {Oussama Harrak and Bruno Casella and Samuele Fonio and Piero Fariselli and Gianluca Mittone and Cesare Rollo and Tiziana Sanavia and Marco Aldinucci},
howpublished = {European Conference on Machine Learning and Principles and Practice of Knowledge Discovery in Databases, 2nd Workshop on Advancements in Federated Learning (WAFL)},
keywords = {icsc, epi, ai, fl},
month = sep,
title = {Federated Adaboost for Survival Analysis},
url = {https://datacloud.di.unito.it/index.php/s/DtXiQfne6BEC235},
year = 2024,
bdsk-url-1 = {https://datacloud.di.unito.it/index.php/s/DtXiQfne6BEC235}} Federated Time Series Classification with ROCKET features
Bruno Casella, Matthias Jakobs, Marco Aldinucci, Sebastian Buschjaeger
icsc
This paper proposes FROCKS, a federated time series classification method using ROCKET features. Our approach dynamically adapts the models' features by selecting and exchanging the best-performing ROCKET kernels from a federation of clients. Specifically, the server gathers the best-performing kernels of the clients together with the associated model parameters, and it performs a weighted average if a kernel is best-performing for more than one client. We compare the proposed method with state-of-the-art approaches on the UCR archive binary classification datasets and show superior performance on most datasets.
@misc{25:casella:frocks,
abstract = {This paper proposes FROCKS, a federated time series classification method using ROCKET features. Our approach dynamically adapts the models' features by selecting and exchanging the best-performing ROCKET kernels from a federation of clients. Specifically, the server gathers the best-performing kernels of the clients together with the associated model parameters, and it performs a weighted average if a kernel is best-performing for more than one client. We compare the proposed method with state-of-the-art approaches on the UCR archive binary classification datasets and show superior performance on most datasets.},
address = {Bruges, Belgium},
author = {Bruno Casella and Matthias Jakobs and Marco Aldinucci and Sebastian Buschjaeger},
howpublished = {32nd European Symposium on Artificial Neural Networks, Computational Intelligence and Machine Learning},
keywords = {icsc},
month = oct,
title = {Federated Time Series Classification with ROCKET features},
url = {https://datacloud.di.unito.it/index.php/s/xWCBsMfjZMkyrAq},
year = 2024,
bdsk-url-1 = {https://datacloud.di.unito.it/index.php/s/xWCBsMfjZMkyrAq}} Towards a European AI Platform
Petr Taborsky, Iacopo Colonnelli, Krzysztof Kurowski, Rakesh Sarma, Niels Henrik Pontoppidan, Branislav Jansík, Nicki Skafte Detlefsen, Jens Egholm Pedersen, Rasmus Larsen, Lars Kai Hansen
ai
@misc{24:colonnelli:eurohpc-user-day,
address = {Amsterdam, Netherlands},
author = {Petr Taborsky and Iacopo Colonnelli and Krzysztof Kurowski and Rakesh Sarma and Niels Henrik Pontoppidan and Branislav Jans\'{i}k and Nicki Skafte Detlefsen and Jens Egholm Pedersen and Rasmus Larsen and Lars Kai Hansen},
howpublished = {2nd EuroHPC User Day},
keywords = {ai},
month = oct,
title = {Towards a {E}uropean {AI} Platform},
url = {https://datacloud.di.unito.it/index.php/s/nbk6YiABGsZXPp4},
year = 2024,
bdsk-url-1 = {https://datacloud.di.unito.it/index.php/s/nbk6YiABGsZXPp4}} FedHP: Federated Learning with Hyperspherical Prototypical Regularization
Samuele Fonio, Mirko Polato, Roberto Esposito
ai fl icsc
This paper introduces FedHP, an innovative algorithm that integrates federated learning, hyperspherical geometries, and prototype learning. Federated Learning (FL) has gained prominence as a privacy- preserving method for building robust models across distributed datasets. Traditionally, FL exchanges model parameters to maintain data privacy; however, in scenarios with expensive data communication, exchanging large neural network models becomes impractical. In such cases, prototype learning offers a viable solution by facilitating the exchange of only a few prototypes. Motivated by these considerations, our approach capitalizes on recent advancements in prototype learning, particularly the advantages offered by non-Euclidean geometries. In addition to presenting FedHP, we offer empirical evidence demonstrating its comparability to other state-of- the-art approaches while significantly reducing communication costs.
@misc{24:esann:fedhp,
abstract = {This paper introduces FedHP, an innovative algorithm that integrates federated learning, hyperspherical geometries, and prototype learning. Federated Learning (FL) has gained prominence as a privacy- preserving method for building robust models across distributed datasets. Traditionally, FL exchanges model parameters to maintain data privacy; however, in scenarios with expensive data communication, exchanging large neural network models becomes impractical. In such cases, prototype learning offers a viable solution by facilitating the exchange of only a few prototypes. Motivated by these considerations, our approach capitalizes on recent advancements in prototype learning, particularly the advantages offered by non-Euclidean geometries. In addition to presenting FedHP, we offer empirical evidence demonstrating its comparability to other state-of- the-art approaches while significantly reducing communication costs.},
address = {Bruges, Belgium},
author = {Samuele Fonio and Mirko Polato and Roberto Esposito},
howpublished = {32nd European Symposium on Artificial Neural Networks, Computational Intelligence and Machine Learning, (ESANN)},
keywords = {icsc, ai, fl},
month = oct,
title = {FedHP: Federated Learning with Hyperspherical Prototypical Regularization},
url = {https://datacloud.di.unito.it/index.php/s/fKyKSSFQKT3LTxW},
year = 2024,
bdsk-url-1 = {https://datacloud.di.unito.it/index.php/s/fKyKSSFQKT3LTxW}} Enabling Cross-Facility LLMs Pre-Training
Gianluca Mittone, Alberto Mulone, Iacopo Colonnelli, Robert Birke, Marco Aldinucci
eupilot fl icsc space streamflow
Big-tech companies pre-train SOTA LLMs on special-purpose, private HPCs, while public research centres lack the resources to compete. We advocate a new take on large model training, e.g., LLMs, called xFFL, which leverages federated learning as an enabling technique to exploit geographically distributed computing power to bridge such digital divide. This work introduces a proof-of-concept federated training of LLaMA-3 8B on three EuroHPC Top500 facilities, proving the viability of leveraging cross-facility publicly available computational power to sustain SOTA LLM workloads.
@misc{24:mittone:TPC,
abstract = {Big-tech companies pre-train SOTA LLMs on special-purpose, private HPCs, while public research centres lack the resources to compete. We advocate a new take on large model training, e.g., LLMs, called xFFL, which leverages federated learning as an enabling technique to exploit geographically distributed computing power to bridge such digital divide. This work introduces a proof-of-concept federated training of LLaMA-3 8B on three EuroHPC Top500 facilities, proving the viability of leveraging cross-facility publicly available computational power to sustain SOTA LLM workloads.},
address = {Atlanta, GE, USA},
author = {Mittone, Gianluca and Mulone, Alberto and Colonnelli, Iacopo and Birke, Robert and Aldinucci, Marco},
howpublished = {Accelerating the Development and Use of Generative AI for Science and Engineering: The Trillion Parameter Consortium ({TPC})},
keywords = {icsc, eupilot, space, fl, streamflow},
month = nov,
title = {Enabling Cross-Facility LLMs Pre-Training},
url = {https://datacloud.di.unito.it/index.php/s/DRgm8ebBkKQgD2d},
year = 2024,
bdsk-url-1 = {https://datacloud.di.unito.it/index.php/s/DRgm8ebBkKQgD2d}} Performance portability via C++ PSTL, SYCL, OpenMP, and HIP: the Gaia AVU-GSR case study
Giulio Malenza
eupex icsc
@misc{24:malenza:p3hpc,
address = {Atlanta, US},
author = {Malenza, Giulio},
howpublished = {{IEEE}},
keywords = {icsc, eupex},
month = nov,
title = {{Performance portability via C++ PSTL, SYCL, OpenMP, and HIP: the Gaia AVU-GSR case study}},
url = {https://datacloud.di.unito.it/index.php/s/55T6LbGcWtssNno},
year = 2024,
bdsk-url-1 = {https://datacloud.di.unito.it/index.php/s/55T6LbGcWtssNno}} Secure workflow computation
Lorenzo Brescia
confidential icsc
@misc{24:brescia:swiss:talk,
address = {Neuchatel, Switzerland},
author = {Brescia, Lorenzo},
howpublished = {{Presentation of previous works and future directions on securing workflows}},
keywords = {icsc, confidential},
month = nov,
title = {{Secure workflow computation}},
url = {https://datacloud.di.unito.it/index.php/s/5nwZ2bi7by3twQB},
year = 2024,
bdsk-url-1 = {https://datacloud.di.unito.it/index.php/s/5nwZ2bi7by3twQB}} 2023
2023
Architecture-Based FedAvg for Vertical Federated Learning
Bruno Casella, Samuele Fonio
Proceedings of the 3rd Workshop on Distributed Machine Learning for the Intelligent Computing Continuum (DML-ICC), IEEE/ACM UCC 2023, Taormina, Italy, 4 December 2023
ai epi fl icsc
Federated Learning (FL) has emerged as a promising solution to address privacy concerns by collaboratively training Deep Learning (DL) models across distributed parties. This work proposes an architecture-based aggregation strategy in Vertical FL, where parties hold data with different attributes but shared instances. Our approach leverages the identical architectural parts, i.e. neural network layers, of different models to selectively aggregate weights, which is particularly relevant when collaborating with institutions holding different types of datasets, i.e., image, text, or tabular datasets. In a scenario where two entities train DL models, such as a Convolutional Neural Network (CNN) and a Multi-Layer Perceptron (MLP), our strategy computes the average only for architecturally identical segments. This preserves data-specific features learned from demographic and clinical data. We tested our approach on two clinical datasets, i.e., the COVID-CXR dataset and the ADNI study. Results show that our method achieves comparable results with the centralized scenario, in which all the data are collected in a single data lake, and benefits from FL generalizability. In particular, compared to the non-federated models, our proposed proof-of-concept model exhibits a slight performance loss on the COVID-CXR dataset (less than 8
@misc{23:casella:architecturalfedavgtalk,
abstract = {Federated Learning (FL) has emerged as a promising solution to address privacy concerns by collaboratively training Deep Learning (DL) models across distributed parties. This work proposes an architecture-based aggregation strategy in Vertical FL, where parties hold data with different attributes but shared instances. Our approach leverages the identical architectural parts, i.e. neural network layers, of different models to selectively aggregate weights, which is particularly relevant when collaborating with institutions holding different types of datasets, i.e., image, text, or tabular datasets. In a scenario where two entities train DL models, such as a Convolutional Neural Network (CNN) and a Multi-Layer Perceptron (MLP), our strategy computes the average only for architecturally identical segments. This preserves data-specific features learned from demographic and clinical data. We tested our approach on two clinical datasets, i.e., the COVID-CXR dataset and the ADNI study. Results show that our method achieves comparable results with the centralized scenario, in which all the data are collected in a single data lake, and benefits from FL generalizability. In particular, compared to the non-federated models, our proposed proof-of-concept model exhibits a slight performance loss on the COVID-CXR dataset (less than 8%), but outperforms ADNI models by up to 12%. Moreover, communication costs between training rounds are minimized by exchanging only the dense layer parameters.},
author = {Casella, Bruno and Fonio, Samuele},
booktitle = {Proceedings of the 3rd Workshop on Distributed Machine Learning for the Intelligent Computing Continuum (DML-ICC), IEEE/ACM UCC 2023, Taormina, Italy, 4 December 2023},
keywords = {icsc, epi, ai, fl},
title = {Architecture-Based FedAvg for Vertical Federated Learning},
url = {https://datacloud.di.unito.it/index.php/s/kJQxnqG4d2ZSicK},
year = 2023,
bdsk-url-1 = {https://datacloud.di.unito.it/index.php/s/kJQxnqG4d2ZSicK}} Towards formal model for location aware workflows
Doriana Medić, Marco Aldinucci
eupex icsc
Designing complex applications and executing them on large-scale topologies of heterogeneous architectures is becoming increasingly crucial in many scientific domains. As a result, diverse workflow modelling paradigms are developed, most of them with no formalisation provided. In these circumstances, comparing two different models or switching from one system to the other becomes a hard nut to crack. This paper investigates the capability of process algebra to model a location aware workflow system. Distributed π-calculus is considered as the base of the formal model due to its ability to describe the communicating components that change their structure as an outcome of the communication. Later, it is discussed how the base model could be extended or modified to capture different features of location aware workflow system. The intention of this paper is to highlight the fact that due to its flexibility, π-calculus, could be a good candidate to represent the behavioural perspective of the workflow system.
@misc{23:wide:medic,
abstract = {Designing complex applications and executing them on large-scale topologies of heterogeneous architectures is becoming increasingly crucial in many scientific domains. As a result, diverse workflow modelling paradigms are developed, most of them with no formalisation provided. In these circumstances, comparing two different models or switching from one system to the other becomes a hard nut to crack. This paper investigates the capability of process algebra to model a location aware workflow system. Distributed $\pi$-calculus is considered as the base of the formal model due to its ability to describe the communicating components that change their structure as an outcome of the communication. Later, it is discussed how the base model could be extended or modified to capture different features of location aware workflow system. The intention of this paper is to highlight the fact that due to its flexibility, $\pi$-calculus, could be a good candidate to represent the behavioural perspective of the workflow system.},
address = {Torino, Italy},
author = {Doriana Medi\'{c} and Marco Aldinucci},
keywords = {eupex, icsc},
title = {Towards formal model for location aware workflows},
url = {https://datacloud.di.unito.it/index.php/s/wpDd9HETzioixTW},
year = 2023,
bdsk-url-1 = {https://datacloud.di.unito.it/index.php/s/wpDd9HETzioixTW}} Federated Learning: A Distributed System Viewpoint
Marco Aldinucci
fl hpc4ai icsc
@misc{23:flw:statale,
address = {Milano, Italy},
author = {Marco Aldinucci},
howpublished = {Universit\`a degli Studi di Milano},
keywords = {icsc, hpc4ai, fl},
month = jan,
title = {Federated Learning: A Distributed System Viewpoint},
year = 2023} From HPC4AI to ICSC living lab: Where systems are the research
Marco Aldinucci
admire eupex eupilot hpc4ai textarossa
@misc{23:Dell:hpc4ai,
address = {Bologna, Italy},
annote = {invited},
author = {Marco Aldinucci},
howpublished = {Dell Advanced Computing Workshop 2023: HPC and Beyond},
keywords = {hpc4ai, eupex, textarossa, admire, eupilot},
month = feb,
title = {From {HPC4AI} to {ICSC} living lab: Where systems are the research},
url = {https://datacloud.di.unito.it/index.php/s/M5QRJyDxyxokcfL},
year = 2023,
bdsk-url-1 = {https://datacloud.di.unito.it/index.php/s/M5QRJyDxyxokcfL}} Paving the way to innovative tools for Federated Learning
Gianluca Mittone
eupilot
Since its debut in 2016, Federated Learning (FL) has been tied to the inner workings of Deep Neural Networks (DNNs). On the one hand, this allowed its development and widespread use as DNNs proliferated. On the other hand, it neglected all those scenarios in which using DNNs is not possible or advantageous. The fact that most current FL frameworks only allow training DNNs reinforces this problem. To address the lack of FL solutions for non-DNN-based use cases, we propose MAFL (Model-Agnostic Federated Learning). MAFL marries a model-agnostic FL algorithm, AdaBoost.F, with an open industry-grade FL framework: Intel® OpenFL. MAFL is the first FL system not tied to any specific type of machine learning model, allowing exploration of FL scenarios beyond DNNs and trees. Furthermore, tools for DML (especially FL) are starting to flourish, many are not flexible and portable enough to experiment with novel systems (e.g., RISC-V), non-fully connected topologies, and asynchronous collaboration schemes. We overcome these limitations via a domain-specific language allowing to map DML schemes to an underlying middleware, i.e. the \ffparallel programming library. As a byproduct, we introduce a RISC-V porting of the PyTorch framework, the first publicly available to our knowledge.
@misc{23:hipeac,
abstract = {Since its debut in 2016, Federated Learning (FL) has been tied to the inner workings of Deep Neural Networks (DNNs). On the one hand, this allowed its development and widespread use as DNNs proliferated. On the other hand, it neglected all those scenarios in which using DNNs is not possible or advantageous. The fact that most current FL frameworks only allow training DNNs reinforces this problem. To address the lack of FL solutions for non-DNN-based use cases, we propose MAFL (Model-Agnostic Federated Learning). MAFL marries a model-agnostic FL algorithm, AdaBoost.F, with an open industry-grade FL framework: Intel{\textregistered} OpenFL. MAFL is the first FL system not tied to any specific type of machine learning model, allowing exploration of FL scenarios beyond DNNs and trees. Furthermore, tools for DML (especially FL) are starting to flourish, many are not flexible and portable enough to experiment with novel systems (e.g., RISC-V), non-fully connected topologies, and asynchronous collaboration schemes. We overcome these limitations via a domain-specific language allowing to map DML schemes to an underlying middleware, i.e. the \ff parallel programming library. As a byproduct, we introduce a RISC-V porting of the PyTorch framework, the first publicly available to our knowledge.},
address = {Toulouse, France},
annote = {invited},
author = {Gianluca Mittone},
howpublished = {2023 HiPEAC Conference},
keywords = {eupilot},
month = feb,
note = {Invited talk},
title = {Paving the way to innovative tools for Federated Learning},
url = {https://datacloud.di.unito.it/index.php/s/2GtxPidHq79RTzA},
year = 2023,
bdsk-url-1 = {https://datacloud.di.unito.it/index.php/s/2GtxPidHq79RTzA}} Standardised Workflows at EBRAINS
Sofia Karvounari, Eleni Mathioulaki, Michael R. Crusoe, Iacopo Colonnelli
across eupex space streamflow
A hands-on training offer for Standardised Workflows in EBRAINS. A short presentation will be used as an introduction, while the main hands-on session will provide information about Writing and Executing Standardised Workflows. TC will give some guidelines, so attendees can experiment with writing CWL tools and workflows and then they will be given access to VM to execute these workflows. The Workflows Dashboard will be also presented during the same session, offering to the attendees the opportunity to understand the different functionalities, use it with TC support and provide useful comments.
@misc{23:HBPSummit,
abstract = {A hands-on training offer for Standardised Workflows in EBRAINS. A short presentation will be used as an introduction, while the main hands-on session will provide information about Writing and Executing Standardised Workflows. TC will give some guidelines, so attendees can experiment with writing CWL tools and workflows and then they will be given access to VM to execute these workflows. The Workflows Dashboard will be also presented during the same session, offering to the attendees the opportunity to understand the different functionalities, use it with TC support and provide useful comments.},
address = {Marseille, France},
annote = {invited},
author = {Sofia Karvounari and Eleni Mathioulaki and Michael R. Crusoe and Iacopo Colonnelli},
howpublished = {Human Brain Project Summit 2023},
keywords = {streamflow, across, eupex, space},
month = mar,
note = {Invited talk},
title = {Standardised Workflows at {EBRAINS}},
url = {https://datacloud.di.unito.it/index.php/s/K5YQKTsX9N7NLT8},
year = 2023,
bdsk-url-1 = {https://datacloud.di.unito.it/index.php/s/K5YQKTsX9N7NLT8}} Experimenting with Systems for Decentralized Machine Learning
Marco Aldinucci
across admire epi eumaster4hpc eupex eupilot hpc4ai space textarossa
Decentralized machine learning (DML) enables collaborative machine learning without centralized input data. Federated learning (FL) and edge inference (EI) are examples of DML. Collaboration naturally happens at the edge of a distributed system with inherently distributed data. While tools for DML are starting to flourish, much needs to be done to get more flexible and portable tools to experiment with novel techniques, non-fully connected topologies, multiple data domains, and asynchronous collaboration schemes. We'll present recent advances in DML, aiming to improve usability in data centers and, at the edge, to widen the class of models extending FL to non-DNN paradigms, to improve the accuracy of models controlling normalization and frequency of communications, and to boost data privacy though generative adversarial networks. Prerequisites: Intermediate understanding of machine learning methods and distributed & parallel computing.
@misc{23:gtc:fl,
abstract = {Decentralized machine learning (DML) enables collaborative machine learning without centralized input data. Federated learning (FL) and edge inference (EI) are examples of DML. Collaboration naturally happens at the edge of a distributed system with inherently distributed data. While tools for DML are starting to flourish, much needs to be done to get more flexible and portable tools to experiment with novel techniques, non-fully connected topologies, multiple data domains, and asynchronous collaboration schemes. We'll present recent advances in DML, aiming to improve usability in data centers and, at the edge, to widen the class of models extending FL to non-DNN paradigms, to improve the accuracy of models controlling normalization and frequency of communications, and to boost data privacy though generative adversarial networks. Prerequisites: Intermediate understanding of machine learning methods and distributed & parallel computing.},
author = {Marco Aldinucci},
howpublished = {NVidia GTC 2023},
keywords = {hpc4ai, eupex, across, textarossa, admire, eupilot, epi, space, eumaster4hpc},
month = mar,
title = {Experimenting with Systems for Decentralized Machine Learning},
url = {https://datacloud.di.unito.it/index.php/s/oyLt7xwkbKxz65c},
year = 2023,
bdsk-url-1 = {https://datacloud.di.unito.it/index.php/s/oyLt7xwkbKxz65c}} HPC4AI: The Research on AI beyond the public cloud
Marco Aldinucci
across admire brainteaser epi eumaster4hpc eupex eupilot hpc4ai space textarossa
@misc{23:CENTAI:hpc4ai,
address = {Torino, Italy},
annote = {invited},
author = {Marco Aldinucci},
howpublished = {CENTAI kick-off meeting},
keywords = {hpc4ai, eupex, across, textarossa, admire, eupilot, epi, eumaster4hpc, space, brainteaser},
month = mar,
title = {{HPC4AI}: The Research on {AI} beyond the public cloud},
url = {https://datacloud.di.unito.it/index.php/s/PZXjPm8sfKTmTGb},
year = 2023,
bdsk-url-1 = {https://datacloud.di.unito.it/index.php/s/PZXjPm8sfKTmTGb}} CWL for HPC: are we there yet?
Iacopo Colonnelli
across eupex streamflow
Modern HPC applications are becoming so heterogeneous and complex that a modular approach to their design, deployment and orchestration is now necessary. This talk explores the benefits of using a vendor-agnostic workflow language (CWL) coupled with a hybrid workflow management system (StreamFlow) in the HPC ecosystem. Also, it will examine the requirements needed to model HPC applications effectively, the CWL's readiness to meet such requirements, and the proposals made to improve the language where needed. Four real use cases will drive the discussion: the ACROSS Project (G.A. n. 955648), where CWL is the primary interface to model three HPC workflows, and the EUPEX Project (G.A. n. 101033975), where StreamFlow will be used for the rapid prototyping of a seismic engineering HPC application for a Modular Supercomputing Architecture (MSA) system.
@misc{23:CWLConference,
abstract = {Modern HPC applications are becoming so heterogeneous and complex that a modular approach to their design, deployment and orchestration is now necessary. This talk explores the benefits of using a vendor-agnostic workflow language (CWL) coupled with a hybrid workflow management system (StreamFlow) in the HPC ecosystem. Also, it will examine the requirements needed to model HPC applications effectively, the CWL's readiness to meet such requirements, and the proposals made to improve the language where needed. Four real use cases will drive the discussion: the ACROSS Project (G.A. n. 955648), where CWL is the primary interface to model three HPC workflows, and the EUPEX Project (G.A. n. 101033975), where StreamFlow will be used for the rapid prototyping of a seismic engineering HPC application for a Modular Supercomputing Architecture (MSA) system.},
address = {Heidelberg, Germany},
annote = {invited},
author = {Iacopo Colonnelli},
howpublished = {2023 CWL Conference},
keywords = {streamflow, across, eupex},
month = mar,
note = {Invited talk},
title = {{CWL} for {HPC}: are we there yet?},
url = {https://datacloud.di.unito.it/index.php/s/CMCd5LiZeXsxwEg},
year = 2023,
bdsk-url-1 = {https://datacloud.di.unito.it/index.php/s/CMCd5LiZeXsxwEg}} Workflows and the Common Workflow Language (CWL)
Iacopo Colonnelli
streamflow
@misc{23:SA2Micro,
address = {Torino, Italy},
author = {Iacopo Colonnelli},
howpublished = {OSA2Micro: An Open Science Approach to Microbiology data integration},
keywords = {streamflow},
month = may,
note = {Invited talk},
title = {Workflows and the Common Workflow Language ({CWL})},
url = {https://datacloud.di.unito.it/index.php/s/NHWWzMMaQgAsA52},
year = 2023,
bdsk-url-1 = {https://datacloud.di.unito.it/index.php/s/NHWWzMMaQgAsA52}} UNITO tools presentation
Iacopo Colonnelli
jupyter-workflow streamflow
@misc{23:FL3WorkingDay,
address = {Bologna, Italy},
author = {Iacopo Colonnelli},
howpublished = {CN HPC Flagship 3 Working Day},
keywords = {streamflow, jupyter-workflow},
month = may,
title = {{UNITO} tools presentation},
url = {https://datacloud.di.unito.it/index.php/s/fgHbnLDQSFtcwLd},
year = 2023,
bdsk-url-1 = {https://datacloud.di.unito.it/index.php/s/fgHbnLDQSFtcwLd}} Experimenting with Emerging RISC-V Systems for Decentralised Machine Learning
Gianluca Mittone, Nicolò Tonci, Robert Birke, Iacopo Colonnelli, Doriana Medić, Andrea Bartolini, Roberto Esposito, Emanuele Parisi, Francesco Beneventi, Mirko Polato, Massimo Torquati, Luca Benini, Marco Aldinucci
ai eupilot icsc
Decentralised Machine Learning (DML) enables collaborative machine learning without centralised input data. Federated Learning (FL) and Edge Inference are examples of DML. While tools for DML (especially FL) are starting to flourish, many are not flexible and portable enough to experiment with novel processors (e.g., RISC-V), non-fully connected network topologies, and asynchronous collaboration schemes. We overcome these limitations via a domain-specific language allowing us to map DML schemes to an underlying middleware, i.e. the FastFlow parallel programming library. We experiment with it by generating different working DML schemes on x86-64 and ARM platforms and an emerging RISC-V one. We characterise the performance and energy efficiency of the presented schemes and systems. As a byproduct, we introduce a RISC-V porting of the PyTorch framework, the first publicly available to our knowledge.
@misc{23:ACMCF,
abstract = {Decentralised Machine Learning (DML) enables collaborative machine learning without centralised input data. Federated Learning (FL) and Edge Inference are examples of DML. While tools for DML (especially FL) are starting to flourish, many are not flexible and portable enough to experiment with novel processors (e.g., RISC-V), non-fully connected network topologies, and asynchronous collaboration schemes. We overcome these limitations via a domain-specific language allowing us to map DML schemes to an underlying middleware, i.e. the FastFlow parallel programming library. We experiment with it by generating different working DML schemes on x86-64 and ARM platforms and an emerging RISC-V one. We characterise the performance and energy efficiency of the presented schemes and systems. As a byproduct, we introduce a RISC-V porting of the PyTorch framework, the first publicly available to our knowledge.},
annote = {invited},
author = {Gianluca Mittone and Nicol{\`o} Tonci and Robert Birke and Iacopo Colonnelli and Doriana Medi{\'c} and Andrea Bartolini and Roberto Esposito and Emanuele Parisi and Francesco Beneventi and Mirko Polato and Massimo Torquati and Luca Benini and Marco Aldinucci},
howpublished = {20th ACM international conference on computing frontiers (CF '23)},
keywords = {eupilot, icsc, ai},
month = may,
note = {Invited talk},
title = {Experimenting with Emerging RISC-V Systems for Decentralised Machine Learning},
url = {https://datacloud.di.unito.it/index.php/s/BYyqZbHzzN4DL8Z},
year = 2023,
bdsk-url-1 = {https://datacloud.di.unito.it/index.php/s/BYyqZbHzzN4DL8Z}} A Federated Learning Benchmark for Drug-Target Interaction
Gianluca Mittone, Filip Svoboda, Marco Aldinucci, Nicholas D. Lane, Pietro Lio'
eupilot icsc
Aggregating pharmaceutical data in the drug-target interaction (DTI) domain can potentially deliver life-saving breakthroughs. It is, however, notoriously difficult due to regulatory constraints and commercial interests. This work proposes the application of federated learning, which is reconcilable with the industry's constraints. It does not require sharing any information that would reveal the entities' data or any other high-level summary. When used on a representative GraphDTA model and the KIBA dataset, it achieves up to 15% improved performance relative to the best available non-privacy preserving alternative. Our extensive battery of experiments shows that, unlike in other domains, the non-IID data distribution in the DTI datasets does not deteriorate FL performance. Additionally, we identify a material trade-off between the benefits of adding new data and the cost of adding more clients.
@misc{23:WWW,
abstract = {Aggregating pharmaceutical data in the drug-target interaction (DTI) domain can potentially deliver life-saving breakthroughs. It is, however, notoriously difficult due to regulatory constraints and commercial interests. This work proposes the application of federated learning, which is reconcilable with the industry's constraints. It does not require sharing any information that would reveal the entities' data or any other high-level summary. When used on a representative GraphDTA model and the KIBA dataset, it achieves up to 15\% improved performance relative to the best available non-privacy preserving alternative. Our extensive battery of experiments shows that, unlike in other domains, the non-IID data distribution in the DTI datasets does not deteriorate FL performance. Additionally, we identify a material trade-off between the benefits of adding new data and the cost of adding more clients.},
annote = {invited},
author = {Gianluca Mittone and Filip Svoboda and Marco Aldinucci and Nicholas D. Lane and Pietro Lio'},
howpublished = {2023 ACM international Web Conference (WWW '23)},
keywords = {eupilot, icsc},
month = may,
note = {Invited talk},
title = {A Federated Learning Benchmark for Drug-Target Interaction},
url = {https://datacloud.di.unito.it/index.php/s/js7go3EorZxSLn9},
year = 2023,
bdsk-url-1 = {https://datacloud.di.unito.it/index.php/s/js7go3EorZxSLn9}} Porting the Variant Calling Pipeline for NGS data in cloud-HPC environment
Alberto Mulone, Sherine Awad, Davide Chiarugi, Marco Aldinucci
across icsc streamflow
In recent years we have understood the importance of analyzing and sequencing human genetic variation. A relevant aspect that emerged from the Covid-19 pandemic was the need to obtain results very quickly; this involved using High-Performance Computing (HPC) environments to execute the Next Generation Sequencing (NGS) pipeline. However, HPC is not always the most suitable environment for the entire execution of a pipeline, especially when it involves many heterogeneous tools. The ability to execute parts of the pipeline on different environments can lead to higher performance but also cheaper executions. This work shows the design and optimization process that led us to a state-of-the-art Variant Calling hybrid workflow based on the StreamFlow Workflow Management System (WfMS). We also compare StreamFlow with Snakemake, an established WfMS targeting HPC facilities, observing comparable performance on single environments and satisfactory improvements with a hybrid cloud-HPC configuration.
@misc{23:mulone:wide:talk,
abstract = {In recent years we have understood the importance of analyzing and sequencing human genetic variation. A relevant aspect that emerged from the Covid-19 pandemic was the need to obtain results very quickly; this involved using High-Performance Computing (HPC) environments to execute the Next Generation Sequencing (NGS) pipeline. However, HPC is not always the most suitable environment for the entire execution of a pipeline, especially when it involves many heterogeneous tools. The ability to execute parts of the pipeline on different environments can lead to higher performance but also cheaper executions. This work shows the design and optimization process that led us to a state-of-the-art Variant Calling hybrid workflow based on the StreamFlow Workflow Management System (WfMS). We also compare StreamFlow with Snakemake, an established WfMS targeting HPC facilities, observing comparable performance on single environments and satisfactory improvements with a hybrid cloud-HPC configuration.},
address = {Torino, Italy},
author = {Alberto Mulone and Sherine Awad and Davide Chiarugi and Marco Aldinucci},
howpublished = {47th {IEEE} Annual Computers, Software, and Applications Conference, {COMPSAC} 2023},
keywords = {across, icsc, streamflow},
month = jun,
title = {Porting the {V}ariant {C}alling {P}ipeline for {NGS} data in cloud-{HPC} environment},
url = {https://datacloud.di.unito.it/index.php/s/zNLj3LCZNsNxHwy},
year = 2023,
bdsk-url-1 = {https://datacloud.di.unito.it/index.php/s/zNLj3LCZNsNxHwy}} Building an accelerated OpenFOAM Proof-of-Concept application using Modern C++.
Giulio Malenza
icsc
@misc{23:OF:genova,
address = {Genova, Italy},
author = {Giulio Malenza},
howpublished = {18th {O}pen{FOAM} Workshop 2023, {G}enova},
keywords = {icsc},
month = jul,
title = {Building an accelerated OpenFOAM Proof-of-Concept application using Modern C++.},
url = {https://datacloud.di.unito.it/index.php/s/mB6omsDB8ERBkGW},
year = 2023,
bdsk-url-1 = {https://datacloud.di.unito.it/index.php/s/mB6omsDB8ERBkGW}} OpenScienceUNITO: Milestone 1: a novel continuous training course for data stewards
Marco Aldinucci, Rossana Damiano
openscience
@misc{23:openscience:cuap,
address = {Torino, Italy},
author = {Marco Aldinucci and Rossana Damiano},
howpublished = {University of Torino research commission},
keywords = {openscience},
month = jul,
note = {OpenScience group of the University of Torino},
title = {OpenScience\@UNITO: Milestone 1: a novel continuous training course for data stewards},
url = {https://datacloud.di.unito.it/index.php/s/r7QgMWTRGF8b2yq},
year = 2023,
bdsk-url-1 = {https://datacloud.di.unito.it/index.php/s/r7QgMWTRGF8b2yq}} Model-Agnostic Federated Learning
Gianluca Mittone, Robert Birke, Marco Aldinucci
eupilot icsc
Since its debut in 2016, Federated Learning (FL) has been tied to the inner workings of Deep Neural Networks (DNNs); this allowed its development as DNNs proliferated but neglected those scenarios in which using DNNs is not possible or advantageous. The fact that most current FL frameworks only support DNNs reinforces this problem. To address the lack of non-DNN-based FL solutions, we propose MAFL (Model-Agnostic Federated Learning). Decentralised Machine Learning (DML) enables collaborative machine learning without centralised input data. Federated Learning (FL) and Edge Inference are examples of DML. While tools for DML (especially FL) are starting to flourish, many are not flexible and portable enough to experiment with novel processors (e.g., RISC-V), non-fully connected network topologies, and asynchronous collaboration schemes. We overcome these limitations via a domain-specific language allowing us to map DML schemes to an underlying middleware, i.e. the FastFlow parallel programming library.
@misc{23:europar:phdtalk,
abstract = {Since its debut in 2016, Federated Learning (FL) has been tied to the inner workings of Deep Neural Networks (DNNs); this allowed its development as DNNs proliferated but neglected those scenarios in which using DNNs is not possible or advantageous. The fact that most current FL frameworks only support DNNs reinforces this problem. To address the lack of non-DNN-based FL solutions, we propose MAFL (Model-Agnostic Federated Learning). Decentralised Machine Learning (DML) enables collaborative machine learning without centralised input data. Federated Learning (FL) and Edge Inference are examples of DML. While tools for DML (especially FL) are starting to flourish, many are not flexible and portable enough to experiment with novel processors (e.g., RISC-V), non-fully connected network topologies, and asynchronous collaboration schemes. We overcome these limitations via a domain-specific language allowing us to map DML schemes to an underlying middleware, i.e. the FastFlow parallel programming library.},
address = {Limassol, Cyprus},
author = {Gianluca Mittone and Robert Birke and Marco Aldinucci},
howpublished = {29th International European Conference on Parallel and Distributed Computing (Euro-Par '23)},
keywords = {eupilot, icsc},
month = aug,
title = {Model-Agnostic Federated Learning},
url = {https://datacloud.di.unito.it/index.php/s/pT3qxkwzzsHR3nS},
year = 2023,
bdsk-url-1 = {https://datacloud.di.unito.it/index.php/s/pT3qxkwzzsHR3nS}} Performance portability in HPC: the Gaia use-case.
Giulio Malenza, Valentina Cesare, Marco Aldinucci
icsc
@misc{23:GAIA:bigHPC,
address = {Naples, Italy},
author = {Malenza, Giulio and Cesare, Valentina and Aldinucci, Marco},
howpublished = {2nd Italian Conference on Big Data and Data Science (ITADATA 2023)},
keywords = {icsc},
month = sep,
title = {Performance portability in {HPC}: the {G}aia use-case.},
url = {https://datacloud.di.unito.it/index.php/s/RqcZpizFtC9toFq},
year = 2023,
bdsk-url-1 = {https://datacloud.di.unito.it/index.php/s/RqcZpizFtC9toFq}} Workflow models for heterogeneous distributed systems
Iacopo Colonnelli
jupyter-workflow streamflow
@misc{23:ITADATABestPhDThesis,
address = {Napoli, Italy},
author = {Iacopo Colonnelli},
howpublished = {2nd Italian Conference on Big Data and Data Science (ITADATA 2023)},
keywords = {streamflow, jupyter-workflow},
month = sep,
note = {Best PhD Thesis Award},
title = {Workflow models for heterogeneous distributed systems},
url = {https://datacloud.di.unito.it/index.php/s/6RqcaJ4djqFNDC8},
year = 2023,
bdsk-url-1 = {https://datacloud.di.unito.it/index.php/s/6RqcaJ4djqFNDC8}} Benchmarking Federated Learning Frameworks for Medical Imaging Tasks
Samuele Fonio
ai eupilot fl icsc
This paper presents a comprehensive benchmarking study of various Federated Learning (FL) frameworks applied to the task of Medical Image Classification. The research specifically addresses the often neglected and complex aspects of scalability and usability in off-the-shelf FL frameworks. Through experimental validation using real case deployments, we provide empirical evidence of the performance and practical relevance of open source FL frameworks. Our findings contribute valuable insights for anyone interested in deploying a FL system, with a particular focus on the healthcare domain—an increasingly attractive field for FL applications.
@misc{23:iciap:benchmed,
abstract = {This paper presents a comprehensive benchmarking study of various Federated Learning (FL) frameworks applied to the task of Medical Image Classification. The research specifically addresses the often neglected and complex aspects of scalability and usability in off-the-shelf FL frameworks. Through experimental validation using real case deployments, we provide empirical evidence of the performance and practical relevance of open source FL frameworks. Our findings contribute valuable insights for anyone interested in deploying a FL system, with a particular focus on the healthcare domain---an increasingly attractive field for FL applications.},
address = {Udine, Italy},
author = {Samuele Fonio},
howpublished = {Image Analysis and Processing - ICIAP 2023 - 22th International Conference - FedMed},
keywords = {icsc, eupilot, ai, fl},
month = sep,
title = {Benchmarking Federated Learning Frameworks for Medical Imaging Tasks},
url = {https://datacloud.di.unito.it/index.php/s/sR7YeTGgfH4DtCR},
year = 2023,
bdsk-url-1 = {https://datacloud.di.unito.it/index.php/s/sR7YeTGgfH4DtCR}} Benchmarking Federated Learning Scalability
Gianluca Mittone, Samuele Fonio
ai eupilot fl icsc
Federated Learning (FL) is a widespread Machine Learning paradigm handling distributed Big Data. In this work, we demonstrate that different FL frameworks expose different scaling performances despite adopting the same technologies, highlighting the need for a more comprehensive study on the topic.
@misc{23:itadata:fl_scaling,
abstract = {Federated Learning (FL) is a widespread Machine Learning paradigm handling distributed Big Data. In this work, we demonstrate that different FL frameworks expose different scaling performances despite adopting the same technologies, highlighting the need for a more comprehensive study on the topic.},
address = {Naples, Italy},
annote = {invited},
author = {Gianluca Mittone and Samuele Fonio},
howpublished = {2nd Italian Conference on Big Data and Data Science (ITADATA 2023)},
keywords = {eupilot, icsc, ai, fl},
month = sep,
title = {Benchmarking Federated Learning Scalability},
url = {https://datacloud.di.unito.it/index.php/s/QZGxC4X3s5LG5oT},
year = 2023,
bdsk-url-1 = {https://datacloud.di.unito.it/index.php/s/QZGxC4X3s5LG5oT}} ACROSS: HPC Big Data Artificial Intelligence Cross Stack Platform Towards Exascale
Iacopo Colonnelli
across streamflow
@misc{23:AssembleaHPC-KTT,
address = {Pisa, Italy},
author = {Iacopo Colonnelli},
howpublished = {LN HPC-KTT Assemblea Nazionale 2023},
keywords = {across, streamflow},
month = oct,
title = {{ACROSS}: {HPC} Big Data Artificial Intelligence Cross Stack Platform Towards Exascale},
url = {https://datacloud.di.unito.it/index.php/s/aK7es8BgFeWorjD},
year = 2023,
bdsk-url-1 = {https://datacloud.di.unito.it/index.php/s/aK7es8BgFeWorjD}} Università degli Studi di Torino: Alpha parallel research group
Iacopo Colonnelli, Doriana Medić, Barbara Cantalupo, Marco Aldinucci
icsc streamflow
@misc{23:HaMMonProject,
address = {Bologna, Italy},
author = {Iacopo Colonnelli and Doriana Medi\'{c} and Barbara Cantalupo and Marco Aldinucci},
howpublished = {HaMMon Kick-Off meeting},
keywords = {icsc, streamflow},
month = oct,
title = {Universit\`{a} degli Studi di Torino: Alpha parallel research group},
url = {https://datacloud.di.unito.it/index.php/s/cmgy9BZ3nwCR2QJ},
year = 2023,
bdsk-url-1 = {https://datacloud.di.unito.it/index.php/s/cmgy9BZ3nwCR2QJ}} Orchestrating Multi-Domain Workflows: The ACROSS Approach
Alberto Scionti, Iacopo Colonnelli
across streamflow
@misc{23:sc:WCIBoF,
address = {Denver, CO, Usa},
author = {Alberto Scionti and Iacopo Colonnelli},
howpublished = {Workflows Community: Modern Workflows for Continuum and Cross-Facility Computing},
keywords = {across, streamflow},
month = nov,
title = {Orchestrating Multi-Domain Workflows: The {ACROSS} Approach},
url = {https://datacloud.di.unito.it/index.php/s/rJXcDBK4mLmS8yz},
year = 2023,
bdsk-url-1 = {https://datacloud.di.unito.it/index.php/s/rJXcDBK4mLmS8yz}} A Systematic Mapping Study of Italian Research on Workflows
Marco Aldinucci, Elena Baralis, Valeria Cardellini, Iacopo Colonnelli, Marco Danelutto, Sergio Decherchi, Giuseppe Di Modica, Luca Ferrucci, Marco Gribaudo, Francesco Iannone, Marco Lapegna, Doriana Medić, Giuseppa Muscianisi, Francesca Righetti, Eva Sciacca, Nicola Tonellotto, Mauro Tortonesi, Paolo Trunfio, Tullio Vardanega
icsc
An entire ecosystem of methodologies and tools revolves around scientific workflow management. They cover crucial non-functional requirements that standard workflow models fail to target, such as interactive execution, energy efficiency, performance portability, Big Data management, and intelligent orchestration in the Computing Continuum. Characterizing and monitoring this ecosystem is crucial to developing an informed view of current and future research directions. This work conducts a systematic mapping study of the Italian workflow research community, analyzing 25 tools and 10 applications from several scientific domains in the context of the “National Research Centre for HPC, Big Data, and Quantum Computing” (ICSC). The study aims to outline the main current research directions and determine how they address the critical needs of modern scientific applications. The findings highlight a variegated research ecosystem of tools, with a prominent interest in advanced workflow orchestration and still immature but promising efforts toward energy efficiency.
@misc{23:sc:works,
abstract = {An entire ecosystem of methodologies and tools revolves around scientific workflow management. They cover crucial non-functional requirements that standard workflow models fail to target, such as interactive execution, energy efficiency, performance portability, Big Data management, and intelligent orchestration in the Computing Continuum. Characterizing and monitoring this ecosystem is crucial to developing an informed view of current and future research directions. This work conducts a systematic mapping study of the Italian workflow research community, analyzing 25 tools and 10 applications from several scientific domains in the context of the ``National Research Centre for HPC, Big Data, and Quantum Computing'' (ICSC). The study aims to outline the main current research directions and determine how they address the critical needs of modern scientific applications. The findings highlight a variegated research ecosystem of tools, with a prominent interest in advanced workflow orchestration and still immature but promising efforts toward energy efficiency.},
address = {Denver, CO, Usa},
author = {Marco Aldinucci and Elena Baralis and Valeria Cardellini and Iacopo Colonnelli and Marco Danelutto and Sergio Decherchi and Giuseppe Di Modica and Luca Ferrucci and Marco Gribaudo and Francesco Iannone and Marco Lapegna and Doriana Medi\'{c} and Giuseppa Muscianisi and Francesca Righetti and Eva Sciacca and Nicola Tonellotto and Mauro Tortonesi and Paolo Trunfio and Tullio Vardanega},
howpublished = {18th Workshop on Workflows in Support of Large-Scale Science (WORKS 2023)},
keywords = {icsc},
month = nov,
title = {A Systematic Mapping Study of Italian Research on Workflows},
url = {https://datacloud.di.unito.it/index.php/s/2kgooG43pGCykji},
year = 2023,
bdsk-url-1 = {https://datacloud.di.unito.it/index.php/s/2kgooG43pGCykji}} Trusted Computing at Scale
Lorenzo Brescia, Iacopo Colonnelli
confidential icsc
@misc{23:brescia:trusted:workflow:fl4:talk,
address = {Turin, Italy},
author = {Brescia, Lorenzo and Colonnelli, Iacopo},
howpublished = {{CN HPC Flagship 4 Working Day}},
keywords = {icsc, confidential},
month = dec,
title = {{Trusted Computing at Scale}},
url = {https://datacloud.di.unito.it/index.php/s/5ij6tLd5SAX4Nn4},
year = 2023,
bdsk-url-1 = {https://datacloud.di.unito.it/index.php/s/5ij6tLd5SAX4Nn4}} The Software Heritage Analytics Framework
Barbara Cantalupo
admire
A framework for analytics on top of Software Heritage
@misc{23:ADMIRE:UG-SHA,
abstract = {A framework for analytics on top of Software Heritage},
address = {Barcelona Supercomputing Centre, Barcelona, Spain},
author = {Cantalupo, Barbara},
howpublished = {Admire User Meeting, 2023},
keywords = {admire},
month = dec,
title = {The {S}oftware {H}eritage {A}nalytics {F}ramework},
url = {https://datacloud.di.unito.it/index.php/s/NbNDPaGZkDc3KyN},
year = 2023,
bdsk-url-1 = {https://datacloud.di.unito.it/index.php/s/NbNDPaGZkDc3KyN}} GreatNector Modelling Tool
Simone Pernice
admire
A tool for system modelling I/O queue behaviour
@misc{23:ADMIRE:UG-GN,
abstract = {A tool for system modelling I/O queue behaviour},
address = {Barcelona Supercomputing Centre, Barcelona, Spain},
author = {Pernice, Simone},
howpublished = {Admire User Meeting, 2023},
keywords = {admire},
month = dec,
title = {Great{N}ector {M}odelling {T}ool},
url = {https://datacloud.di.unito.it/index.php/s/NbNDPaGZkDc3KyN},
year = 2023,
bdsk-url-1 = {https://datacloud.di.unito.it/index.php/s/NbNDPaGZkDc3KyN}} Distributed Edge Inference: an Experimental Study on Multiview Detection
Gianluca Mittone, Giulio Malenza, Marco Aldinucci, Robert Birke
ai eupilot icsc
Computing is evolving rapidly to cater to the increasing demand for sophisticated services, and Cloud computing lays a solid foundation for flexible on-demand provisioning. However, as the size of applications grows, the centralised client-server approach used by Cloud computing increasingly limits the applications scalability. To achieve ultra-scalability, cloud/edge/fog computing converges into the compute continuum, completely decentralising the infrastructure to encompass universal, pervasive resources. The compute continuum makes devising applications benefitting from this complex environment a challenging research problem. We put the opportunities the compute continuum others to the test through a real-world multi-view detection model (MvDet) implemented with the FastFL C/C++ high-performance edge inference framework. Computational performance is discussed considering many experimental scenarios, encompassing different edge computational capabilities and network bandwidths. We obtain up to 1.92x speedup in inference time over a centralised solution using the same devices.
@misc{23:ucc:multiview,
abstract = {Computing is evolving rapidly to cater to the increasing demand for sophisticated services, and Cloud computing lays a solid foundation for flexible on-demand provisioning. However, as the size of applications grows, the centralised client-server approach used by Cloud computing increasingly limits the applications scalability. To achieve ultra-scalability, cloud/edge/fog computing converges into the compute continuum, completely decentralising the infrastructure to encompass universal, pervasive resources. The compute continuum makes devising applications benefitting from this complex environment a challenging research problem. We put the opportunities the compute continuum others to the test through a real-world multi-view detection model (MvDet) implemented with the FastFL C/C++ high-performance edge inference framework. Computational performance is discussed considering many experimental scenarios, encompassing different edge computational capabilities and network bandwidths. We obtain up to 1.92x speedup in inference time over a centralised solution using the same devices.},
address = {Taormina, Italy},
annote = {invited},
author = {Gianluca Mittone and Giulio Malenza and Marco Aldinucci and Robert Birke},
howpublished = {The 16th IEEE/ACM International Conference on Utility and Cloud Computing (UCC 2023)},
keywords = {eupilot, icsc, ai},
month = dec,
title = {Distributed Edge Inference: an Experimental Study on Multiview Detection},
url = {https://datacloud.di.unito.it/index.php/s/XfjNZEPSNfSKPFr},
year = 2023,
bdsk-url-1 = {https://datacloud.di.unito.it/index.php/s/XfjNZEPSNfSKPFr}} 2022
2022
The OpenDeepHealth toolkit
Iacopo Colonnelli
deephealth hpc4ai
@misc{22:DHWinterSchool,
address = {Torino, Italy},
author = {Iacopo Colonnelli},
howpublished = {DeepHealth Winter School},
keywords = {deephealth, hpc4ai},
month = jan,
title = {The {OpenDeepHealth} toolkit},
url = {https://datacloud.di.unito.it/index.php/s/cJ8pRNsWRrfwPqr},
year = 2022,
bdsk-url-1 = {https://datacloud.di.unito.it/index.php/s/cJ8pRNsWRrfwPqr}} Transfer Learning via Test-Time Neural Networks Aggregation
Bruno Casella
Presentation of the paper "Transfer Learning via Test-Time Neural Networks Aggregation" to VISAPP, the 19th International Joint Conference on Computer Vision, Imaging and Computer Graphics Theory and Applications
@misc{22:visapp,
abstract = {Presentation of the paper "Transfer Learning via Test-Time Neural Networks Aggregation" to VISAPP, the 19th International Joint Conference on Computer Vision, Imaging and Computer Graphics Theory and Applications},
address = {Virtual event},
author = {Bruno Casella},
howpublished = {VISAPP},
month = feb,
title = {Transfer Learning via Test-Time Neural Networks Aggregation},
url = {https://datacloud.di.unito.it/index.php/s/AKMT87eZDFX46JJ},
year = 2022,
bdsk-url-1 = {https://datacloud.di.unito.it/index.php/s/AKMT87eZDFX46JJ}} StreamFlow: A framework for hybrid workflows
Iacopo Colonnelli
across streamflow
@misc{22:across-streamflow,
address = {Virtual event},
author = {Iacopo Colonnelli},
howpublished = {ACROSS WP4 meeting},
keywords = {streamflow, across},
month = feb,
title = {{StreamFlow}: A framework for hybrid workflows},
url = {https://datacloud.di.unito.it/index.php/s/FXFTKtQSRf6anMX},
year = 2022,
bdsk-url-1 = {https://datacloud.di.unito.it/index.php/s/FXFTKtQSRf6anMX}} Poster Session VISIGRAPP 2022
Bruno Casella
@misc{22:VISIGRAPP,
address = {Torino, Italy},
annote = {invited},
author = {Bruno Casella},
date-added = {2022-02-07 14:00:00 +0100},
howpublished = {VISIGRAPP 2022},
month = feb,
note = {Session chair},
title = {Poster Session VISIGRAPP 2022},
year = {2022},
bdsk-url-1 = {}} Towards Machine Learning on the edge: the opportunities of Federated Learning
Bruno Casella, Gianluca Mittone, Yasir Arfat, Marco Aldinucci, Roberto Esposito
ai fl
Since its debut in 2016, Federated Learning (FL) has been tied to the inner workings of Deep Neural Networks (DNNs). On the one hand, this allowed its development and widespread use as DNNs proliferated. On the other hand, it neglected all those scenarios in which using DNNs is not possible or advantageous. The fact that most current FL frameworks only allow training DNNs reinforces this problem. To address the lack of FL solutions for non-DNN-based use cases, we propose MAFL (Model-Agnostic Federated Learning). MAFL marries a model-agnostic FL algorithm, AdaBoost.F, with an open industry-grade FL framework: Intel® OpenFL. MAFL is the first FL system not tied to any specific type of machine learning model, allowing exploration of FL scenarios beyond DNNs and trees. Furthermore, tools for DML (especially FL) are starting to flourish, many are not flexible and portable enough to experiment with novel systems (e.g., RISC-V), non-fully connected topologies, and asynchronous collaboration schemes. We overcome these limitations via a domain-specific language allowing to map DML schemes to an underlying middleware, i.e. the \ffparallel programming library.
@misc{22:ITWSHPC:FL,
abstract = {Since its debut in 2016, Federated Learning (FL) has been tied to the inner workings of Deep Neural Networks (DNNs). On the one hand, this allowed its development and widespread use as DNNs proliferated. On the other hand, it neglected all those scenarios in which using DNNs is not possible or advantageous. The fact that most current FL frameworks only allow training DNNs reinforces this problem. To address the lack of FL solutions for non-DNN-based use cases, we propose MAFL (Model-Agnostic Federated Learning). MAFL marries a model-agnostic FL algorithm, AdaBoost.F, with an open industry-grade FL framework: Intel{\textregistered} OpenFL. MAFL is the first FL system not tied to any specific type of machine learning model, allowing exploration of FL scenarios beyond DNNs and trees. Furthermore, tools for DML (especially FL) are starting to flourish, many are not flexible and portable enough to experiment with novel systems (e.g., RISC-V), non-fully connected topologies, and asynchronous collaboration schemes. We overcome these limitations via a domain-specific language allowing to map DML schemes to an underlying middleware, i.e. the \ff parallel programming library.},
address = {Turin, Italy},
annote = {invited},
author = {Bruno Casella and Gianluca Mittone and Yasir Arfat and Marco Aldinucci and Roberto Esposito},
date-modified = {2026-09-13 23:04:13 +0200},
howpublished = {3rd Italian Workshop on HPC},
keywords = {fl,ai},
month = sep,
note = {Invited talk},
title = {Towards Machine Learning on the edge: the opportunities of Federated Learning},
url = {https://datacloud.di.unito.it/index.php/s/X3bDLCBeZMo4qr5},
year = {2022},
bdsk-url-1 = {https://datacloud.di.unito.it/index.php/s/X3bDLCBeZMo4qr5}} Python for the human heart: a case study
Gianluca Mittone
Presentation of the "Machine learning-based prediction of adverse events following an acute coronary syndrome (PRAISE): a modelling study of pooled datasets" paper at the PyCon22 DataBeer event.
@misc{22:PyCon,
abstract = {Presentation of the "Machine learning-based prediction of adverse events following an acute coronary syndrome (PRAISE): a modelling study of pooled datasets" paper at the PyCon22 DataBeer event.},
address = {Florence, Italy},
annote = {invited},
author = {Gianluca Mittone},
howpublished = {2022 Python Conference},
month = may,
note = {Invited talk},
title = {Python for the human heart: a case study},
url = {https://datacloud.di.unito.it/index.php/s/fERJB5xeE6cXm4N},
year = {2022},
bdsk-url-1 = {https://datacloud.di.unito.it/index.php/s/fERJB5xeE6cXm4N}} Benchmarking FedAvg and FedCurv for Image Classification Tasks
Bruno Casella
eupilot
Presentation of the paper "Benchmarking FedAvg and FedCurv for Image Classification Tasks" to the first italian conference on Big Data and Data Science
@misc{22:itadata,
abstract = {Presentation of the paper "Benchmarking FedAvg and FedCurv for Image Classification Tasks" to the first italian conference on Big Data and Data Science},
address = {Milan, Italy},
author = {Bruno Casella},
howpublished = {ITADATA},
keywords = {eupilot},
month = sep,
title = {Benchmarking FedAvg and FedCurv for Image Classification Tasks},
url = {https://datacloud.di.unito.it/index.php/s/6XaEXnAowRrAHGL},
year = {2022},
bdsk-url-1 = {https://datacloud.di.unito.it/index.php/s/6XaEXnAowRrAHGL}} Il calcolo parallelo: una storia di metodi e algoritmi raccontata dalle macchine
Marco Aldinucci
across admire eumaster4hpc eupex eupilot textarossa
Lectio Magistralis alle finali nazionali delle Olimpiadi di Informatica 2022
@misc{22:olimpiadi:cs,
abstract = {Lectio Magistralis alle finali nazionali delle Olimpiadi di Informatica 2022},
address = {Biella, Italy},
annote = {invited},
author = {Marco Aldinucci},
howpublished = {Olimpiadi di Informatica},
keywords = {eupex, across, textarossa, admire, eupilot, eumaster4hpc},
month = sep,
note = {Invited talk},
title = {Il calcolo parallelo: una storia di metodi e algoritmi raccontata dalle macchine},
url = {https://datacloud.di.unito.it/index.php/s/7ZdfLkn3NetzXCN},
year = {2022},
bdsk-url-1 = {https://datacloud.di.unito.it/index.php/s/7ZdfLkn3NetzXCN}} La convergenza HPC-cloud è l'anello mancante tra il calcolo scientifico e l'IA applicata
Marco Aldinucci
across admire eumaster4hpc eupex eupilot textarossa
Innanzitutto, le infrastrutture HPC stanno adottando le GPU per il loro rapporto prestazioni per watt superiore rispetto ai multicore generici. In secondo luogo, i flussi di lavoro scientifici di prossima generazione stanno integrando passaggi basati sull'intelligenza artificiale per la loro precisione nell'approssimazione e nell'analisi di fenomeni complessi. In terzo luogo, l'IA e in particolare il Machine Learning (ML) rappresentano un carico di lavoro perfetto per le GPU in termini di prestazioni e tempo di sviluppo. Oggi non possiamo ancora chiudere il cerchio eseguendo senza problemi carichi di lavoro scientifici abilitati all'intelligenza artificiale nelle infrastrutture HPC perché il loro software di sistema e gli strumenti di sviluppo non sono progettati per i carichi di lavoro moderni, come i framework ML progettati per il cloud. È probabile che la convergenza HPC-cloud colmi il divario. Nel talk verranno presentate le infrastrutture e gli strumenti sviluppati all'Università di Torino per la convergenza HPC-cloud (es. HPC4AI, StreamFlow, CAPIO, Jupyter-workflow) e come sono stati utilizzati per le applicazioni di intelligenza artificiale, come la diagnosi spiegabile di polmonite COVID-19 e la tutela della privacy AI. L'esperienza maturata nella progettazione e gestione di HPC4AI costituisce il cuore della progettazione del laboratorio di contaminazione del "FutureHPC" di Torino secondo il Centro Nazionale "HPC, BigData e Quantum Computing" finanziato dal PNRR con 320M€ che dovrebbe essere operativo dal 1 settembre 2022. L'obiettivo finale del laboratorio di contaminazione è sviluppare relazioni e collaborazioni tra industria e università.
@misc{22:soiel:ai,
abstract = {Innanzitutto, le infrastrutture HPC stanno adottando le GPU per il loro rapporto prestazioni per watt superiore rispetto ai multicore generici. In secondo luogo, i flussi di lavoro scientifici di prossima generazione stanno integrando passaggi basati sull'intelligenza artificiale per la loro precisione nell'approssimazione e nell'analisi di fenomeni complessi. In terzo luogo, l'IA e in particolare il Machine Learning (ML) rappresentano un carico di lavoro perfetto per le GPU in termini di prestazioni e tempo di sviluppo. Oggi non possiamo ancora chiudere il cerchio eseguendo senza problemi carichi di lavoro scientifici abilitati all'intelligenza artificiale nelle infrastrutture HPC perch{\'e} il loro software di sistema e gli strumenti di sviluppo non sono progettati per i carichi di lavoro moderni, come i framework ML progettati per il cloud. {\`E} probabile che la convergenza HPC-cloud colmi il divario. Nel talk verranno presentate le infrastrutture e gli strumenti sviluppati all'Universit{\`a} di Torino per la convergenza HPC-cloud (es. HPC4AI, StreamFlow, CAPIO, Jupyter-workflow) e come sono stati utilizzati per le applicazioni di intelligenza artificiale, come la diagnosi spiegabile di polmonite COVID-19 e la tutela della privacy AI. L'esperienza maturata nella progettazione e gestione di HPC4AI costituisce il cuore della progettazione del laboratorio di contaminazione del "FutureHPC" di Torino secondo il Centro Nazionale "HPC, BigData e Quantum Computing" finanziato dal PNRR con 320M€ che dovrebbe essere operativo dal 1 settembre 2022. L'obiettivo finale del laboratorio di contaminazione {\`e} sviluppare relazioni e collaborazioni tra industria e universit{\`a}.},
address = {Virtual event},
annote = {invited},
author = {Marco Aldinucci},
howpublished = {Intelligenza Artificiale e Business Applications},
keywords = {eupex, across, textarossa, admire, eupilot, eumaster4hpc},
month = sep,
note = {Invited talk},
title = {La convergenza HPC-cloud {\`e} l'anello mancante tra il calcolo scientifico e l'IA applicata},
url = {https://datacloud.di.unito.it/index.php/s/xCQSqJ8bCKCXMK9},
year = {2022},
bdsk-url-1 = {https://datacloud.di.unito.it/index.php/s/xCQSqJ8bCKCXMK9}} Hybrid Workflows For Large-Scale Scientific Applications
Iacopo Colonnelli, Marco Aldinucci
across eupex jupyter-workflow textarossa
Large-scale scientific applications are facing an irreversible transition from monolithic, high-performance oriented codes to modular and polyglot deployments of specialised (micro-)services. The reasons behind this transition are many: coupling of standard solvers with Deep Learning techniques, offloading of data analysis and visualisation to Cloud, and the advent of specialised hardware accelerators. Topology-aware Workflow Management Systems (WMSs) play a crucial role. In particular, topology-awareness allows an explicit mapping of workflow steps onto heterogeneous locations, allowing automated executions on top of hybrid architectures (e.g., cloud+HPC or classical+quantum). Plus, topology-aware WMSs can offer non-functional requirements OOTB, e.g. components' life-cycle orchestration, secure and efficient data transfers, fault tolerance, and cross-cluster execution of urgent workloads. Augmenting interactive Jupyter Notebooks with distributed workflow capabilities allows domain experts to prototype and scale applications using the same technological stack, while relying on a feature-rich and user-friendly web interface. This abstract will showcase how these general methodologies can be applied to a typical geoscience simulation pipeline based on the Full Wavefront Inversion (FWI) technique. In particular, a prototypical Jupyter Notebook will be executed interactively on Cloud. Preliminary data analyses and post-processing will be executed locally, while the computationally demanding optimisation loop will be scheduled on a remote HPC cluster.
@misc{22:eage,
abstract = {Large-scale scientific applications are facing an irreversible transition from monolithic, high-performance oriented codes to modular and polyglot deployments of specialised (micro-)services. The reasons behind this transition are many: coupling of standard solvers with Deep Learning techniques, offloading of data analysis and visualisation to Cloud, and the advent of specialised hardware accelerators. Topology-aware Workflow Management Systems (WMSs) play a crucial role. In particular, topology-awareness allows an explicit mapping of workflow steps onto heterogeneous locations, allowing automated executions on top of hybrid architectures (e.g., cloud+HPC or classical+quantum). Plus, topology-aware WMSs can offer non-functional requirements OOTB, e.g. components' life-cycle orchestration, secure and efficient data transfers, fault tolerance, and cross-cluster execution of urgent workloads. Augmenting interactive Jupyter Notebooks with distributed workflow capabilities allows domain experts to prototype and scale applications using the same technological stack, while relying on a feature-rich and user-friendly web interface. This abstract will showcase how these general methodologies can be applied to a typical geoscience simulation pipeline based on the Full Wavefront Inversion (FWI) technique. In particular, a prototypical Jupyter Notebook will be executed interactively on Cloud. Preliminary data analyses and post-processing will be executed locally, while the computationally demanding optimisation loop will be scheduled on a remote HPC cluster.},
address = {Milano, Italy},
author = {Iacopo Colonnelli and Marco Aldinucci},
howpublished = {6th EAGE High Performance Computing Workshop},
keywords = {eupex, across, textarossa, jupyter-workflow},
month = sep,
title = {Hybrid Workflows For Large-Scale Scientific Applications},
url = {https://datacloud.di.unito.it/index.php/s/GScPS5LCPdt6Yoo},
year = {2022},
bdsk-url-1 = {https://datacloud.di.unito.it/index.php/s/GScPS5LCPdt6Yoo}} Hybrid workflows for heterogeneous distributed computing
Iacopo Colonnelli, Barbara Cantalupo, Doriana Medić, Marco Aldinucci
across admire eumaster4hpc eupex eupilot textarossa
@misc{22:itwshpc,
address = {Torino, Italy},
author = {Iacopo Colonnelli and Barbara Cantalupo and Doriana Medi\'{c} and Marco Aldinucci},
howpublished = {3rd Italian Workshop on HPC (ITWSHPC)},
keywords = {eupex, across, admire, eupilot, textarossa, eumaster4hpc},
month = sep,
title = {Hybrid workflows for heterogeneous distributed computing},
url = {https://datacloud.di.unito.it/index.php/s/ienbcA2DJ26aioE},
year = {2022},
bdsk-url-1 = {https://datacloud.di.unito.it/index.php/s/ienbcA2DJ26aioE}} CINI HPC-KTT: HPC Key Technologies and Tools National Lab
Iacopo Colonnelli, Marco Aldinucci
across admire eumaster4hpc eupex eupilot textarossa
@misc{22:nvidia_hpc_roundtable,
address = {Casalecchio di Reno, Italy},
annote = {invited},
author = {Iacopo Colonnelli and Marco Aldinucci},
howpublished = {NVIDIA HPC Roundtable},
keywords = {eupex, across, admire, eupilot, textarossa, eumaster4hpc},
month = sep,
note = {Invited talk},
title = {{CINI HPC-KTT}: {HPC} {K}ey {T}echnologies and {T}ools National Lab},
url = {https://datacloud.di.unito.it/index.php/s/9EQniZ2dGzdJ26f},
year = {2022},
bdsk-url-1 = {https://datacloud.di.unito.it/index.php/s/9EQniZ2dGzdJ26f}} Cognitive continuum: a game theoretical approach
Marco Aldinucci
across admire brainteaser eumaster4hpc eupex eupilot textarossa
Cognitive continuum: a game theoretical approach, (maybe) data operations are too basic: read, write, copy, remove … The talk is aimed to contribute to the forthcoming HiPEAC Vision document
@misc{22:hipeacvision:fl,
abstract = {Cognitive continuum: a game theoretical approach, (maybe) data operations are too basic: read, write, copy, remove {\ldots} The talk is aimed to contribute to the forthcoming HiPEAC Vision document},
address = {Brussels, Belgium},
annote = {invited},
author = {Marco Aldinucci},
howpublished = {HiPEAC Vision meeting, Brussels, 16 May 2022},
keywords = {admire, eupex, across, eupilot, textarossa, eumaster4hpc, brainteaser},
month = may,
title = {Cognitive continuum: a game theoretical approach},
url = {https://datacloud.di.unito.it/index.php/s/453HWfmrQyo7j9E},
year = {2022},
bdsk-url-1 = {https://datacloud.di.unito.it/index.php/s/453HWfmrQyo7j9E}} EuroHPC and the Italian HPC ecosystem
Marco Aldinucci
across admire eumaster4hpc eupex eupilot icsc textarossa
The talk presents the main investments currently ongoing in Italy in the HPC area as well as the activity of Italian stakeholders within EuroHPC. The novel Italian National Centre on HPC (ICSC) is introduced.
@misc{22:cip:romania,
abstract = {The talk presents the main investments currently ongoing in Italy in the HPC area as well as the activity of Italian stakeholders within EuroHPC. The novel Italian National Centre on HPC (ICSC) is introduced.},
address = {Bucharest, Romania},
annote = {invited},
author = {Marco Aldinucci},
howpublished = {Critical Infrastructure Protection Forum - EuroCC Romania},
keywords = {admire, eupex, across, eupilot, textarossa, eumaster4hpc, icsc},
month = jun,
note = {Invited talk},
title = {EuroHPC and the Italian HPC ecosystem},
url = {https://datacloud.di.unito.it/index.php/s/5dFFoNsZzwTzQkn},
year = {2022},
bdsk-url-1 = {https://datacloud.di.unito.it/index.php/s/5dFFoNsZzwTzQkn}} The Italian HPC ecosystem and the next generation of EuroHPC CoE
Marco Aldinucci
across admire eumaster4hpc eupex eupilot icsc textarossa
The talk presents the main investments currently ongoing in Italy in the HPC area as well as the activity of Italian stakeholders within EuroHPC. The novel Italian National Centre on HPC (ICSC) is introduced.
@misc{22:eocoe:summit,
abstract = {The talk presents the main investments currently ongoing in Italy in the HPC area as well as the activity of Italian stakeholders within EuroHPC. The novel Italian National Centre on HPC (ICSC) is introduced.},
address = {Napoli, Italy},
annote = {invited},
author = {Marco Aldinucci},
howpublished = {EuroHPC EoCoE final summit},
keywords = {admire, eupex, across, eupilot, textarossa, eumaster4hpc, icsc},
month = jun,
note = {Invited talk},
title = {The Italian HPC ecosystem and the next generation of EuroHPC CoE},
url = {https://datacloud.di.unito.it/index.php/s/AH5Ms3NekeoEooB},
year = {2022},
bdsk-url-1 = {https://datacloud.di.unito.it/index.php/s/AH5Ms3NekeoEooB}} Dossier: multi-tenant distributed Jupyter Notebooks
Iacopo Colonnelli, Dario Tranchitella
across deephealth hpc4ai jupyter-workflow
When providing data analysis as a service, one must tackle several problems. Data privacy and protection by design are crucial when working on sensitive data. Performance and scalability are fundamental for compute-intensive workloads, e.g. training Deep Neural Networks. User-friendly interfaces and fast prototyping tools are essential to allow domain experts to experiment with new techniques. Portability and reproducibility are necessary to assess the actual value of results. Kubernetes is the best platform to provide reliable, elastic, and maintainable services. However, Kubernetes alone is not enough to achieve large-scale multi-tenant reproducible data analysis. OOTB support for multi-tenancy is too rough, with only two levels of segregation (i.e. the single namespace or the entire cluster). Offloading computation to off-cluster resources is non-trivial and requires the user's manual configuration. Also, Jupyter Notebooks per se cannot provide much scalability (they execute locally and sequentially) and reproducibility (users can run cells in any order and any number of times). The Dossier platform allows system administrators to manage multi-tenant distributed Jupyter Notebooks at the cluster level in the Kubernetes way, i.e. through CRDs. Namespaces are aggregated in Tenants, and all security and accountability aspects are managed at that level. Each Notebook spawns into a user-dedicated namespace, subject to all Tenant-level constraints. Users can rely on provisioned resources, either in-cluster worker nodes or external resources like HPC facilities. Plus, they can plug their computing nodes in a BYOD fashion. Notebooks are interpreted as distributed workflows, where each cell is a task that one can offload to a different location in charge of its execution.
@misc{22:data-on-kubernetes,
abstract = {When providing data analysis as a service, one must tackle several problems. Data privacy and protection by design are crucial when working on sensitive data. Performance and scalability are fundamental for compute-intensive workloads, e.g. training Deep Neural Networks. User-friendly interfaces and fast prototyping tools are essential to allow domain experts to experiment with new techniques. Portability and reproducibility are necessary to assess the actual value of results. Kubernetes is the best platform to provide reliable, elastic, and maintainable services. However, Kubernetes alone is not enough to achieve large-scale multi-tenant reproducible data analysis. OOTB support for multi-tenancy is too rough, with only two levels of segregation (i.e. the single namespace or the entire cluster). Offloading computation to off-cluster resources is non-trivial and requires the user's manual configuration. Also, Jupyter Notebooks per se cannot provide much scalability (they execute locally and sequentially) and reproducibility (users can run cells in any order and any number of times). The Dossier platform allows system administrators to manage multi-tenant distributed Jupyter Notebooks at the cluster level in the Kubernetes way, i.e. through CRDs. Namespaces are aggregated in Tenants, and all security and accountability aspects are managed at that level. Each Notebook spawns into a user-dedicated namespace, subject to all Tenant-level constraints. Users can rely on provisioned resources, either in-cluster worker nodes or external resources like HPC facilities. Plus, they can plug their computing nodes in a BYOD fashion. Notebooks are interpreted as distributed workflows, where each cell is a task that one can offload to a different location in charge of its execution.},
address = {Virtual event},
author = {Iacopo Colonnelli and Dario Tranchitella},
howpublished = {DoK Talks 141},
keywords = {jupyter-workflow, across, deephealth, hpc4ai},
month = jul,
note = {Invited talk},
title = {Dossier: multi-tenant distributed {J}upyter {N}otebooks},
url = {https://datacloud.di.unito.it/index.php/s/RNqTGmTqWS66qHT},
year = {2022},
bdsk-url-1 = {https://datacloud.di.unito.it/index.php/s/RNqTGmTqWS66qHT}} StreamFlow
Iacopo Colonnelli
across deephealth eupex streamflow textarossa
@misc{22:healthycloud-workshop,
address = {Virtual event},
annote = {invited},
author = {Iacopo Colonnelli},
howpublished = {2nd HealthyCloud Workshop: Analysis of existing orchestration mechanisms for distributed computational analyses},
keywords = {streamflow, deephealth, across, eupex, textarossa},
month = jul,
note = {Invited talk},
title = {{StreamFlow}},
url = {https://datacloud.di.unito.it/index.php/s/Taz8qtzmkmn9ffT},
year = {2022},
bdsk-url-1 = {https://datacloud.di.unito.it/index.php/s/Taz8qtzmkmn9ffT}} T4.1: Streaming models
Iacopo Colonnelli, Marco Aldinucci
textarossa
@misc{22:textarossa-ga-meeting,
address = {Roma, Italy},
author = {Iacopo Colonnelli and Marco Aldinucci},
howpublished = {TEXTAROSSA General Meeting},
keywords = {textarossa},
month = jun,
title = {T4.1: Streaming models},
url = {https://datacloud.di.unito.it/index.php/s/cNBnwSnTc8GiCkN},
year = {2022},
bdsk-url-1 = {https://datacloud.di.unito.it/index.php/s/cNBnwSnTc8GiCkN}} StreamFlow: a topology-aware WMS
Iacopo Colonnelli
across deephealth eupex streamflow textarossa
@misc{22:elixir-streamflow,
address = {Virtual event},
annote = {invited},
author = {Iacopo Colonnelli},
howpublished = {ELIXIR Cloud, Data & AAI Bi-weekly Technical Calls},
keywords = {streamflow, deephealth, across, eupex, textarossa},
month = jun,
note = {Invited talk},
title = {{StreamFlow}: a topology-aware {WMS}},
url = {https://datacloud.di.unito.it/index.php/s/Z9GsKnRCxmBdMd3},
year = {2022},
bdsk-url-1 = {https://datacloud.di.unito.it/index.php/s/Z9GsKnRCxmBdMd3}} From small files to no files
Marco Aldinucci
admire eupex
Modern distributed high-performance storage systems saturate the network bandwidth, and the margins for improvement at the software level are tiny. Due to metadata access, they might be troubled with massive access to small files. An example is the Software Heritage (SH) dataset, half petabytes of files with an average size of 3kBytes (Terabytes of metadata). While working with SH, we developed the idea of substituting files with in-memory streams. We did it living in dread with the fear of asking application programmers to rewrite their lovely antique legacy code exploiting the POSIX interface, and up to now, we did not. In the talk, we will introduce CAPIO (Cross-Application Programmable I/O) design principles and the current state of development of the prototype.
@misc{22:p3s:capio,
abstract = {Modern distributed high-performance storage systems saturate the network bandwidth, and the margins for improvement at the software level are tiny. Due to metadata access, they might be troubled with massive access to small files. An example is the Software Heritage (SH) dataset, half petabytes of files with an average size of 3kBytes (Terabytes of metadata). While working with SH, we developed the idea of substituting files with in-memory streams. We did it living in dread with the fear of asking application programmers to rewrite their lovely antique legacy code exploiting the POSIX interface, and up to now, we did not. In the talk, we will introduce CAPIO (Cross-Application Programmable I/O) design principles and the current state of development of the prototype.},
address = {Paris, France},
annote = {https://per3s.github.io, invited},
author = {Marco Aldinucci},
howpublished = {6th Workshop on Performance and Scalability of Storage Systems},
keywords = {admire, eupex},
month = jun,
note = {Invited talk},
title = {From small files to no files},
url = {https://datacloud.di.unito.it/index.php/s/KLDi87xQmX86iXg},
year = {2022},
bdsk-url-1 = {https://datacloud.di.unito.it/index.php/s/KLDi87xQmX86iXg}} HPC-cloud convergence is the missing link between scientific computing and applied-AI
Marco Aldinucci
across deephealth eupex eupilot
First, HPC infrastructures are embracing GPUs for their superior performance-per-watt ratio against general-purpose multicores. Second, the next-generation scientific workflows are integrating AI-based steps for their accuracy in approximating and analyzing complex phenomena. Third, AI and specifically Machine Learning (ML), is a perfect workload for GPUs in terms of performance and development time. Today, we cannot still close the circle seamlessly running AI-enabled scientific workloads into HPC infrastructures because their system software and development tools are not designed for modern workloads, such as ML frameworks designed for the cloud. HPC-cloud convergence is likely to bridge the gap. In the talk, we will present Streamflow and CAPIO, two development tools for HPC-cloud convergence.
@misc{22:ml4astrotalk,
abstract = {First, HPC infrastructures are embracing GPUs for their superior performance-per-watt ratio against general-purpose multicores. Second, the next-generation scientific workflows are integrating AI-based steps for their accuracy in approximating and analyzing complex phenomena. Third, AI and specifically Machine Learning (ML), is a perfect workload for GPUs in terms of performance and development time. Today, we cannot still close the circle seamlessly running AI-enabled scientific workloads into HPC infrastructures because their system software and development tools are not designed for modern workloads, such as ML frameworks designed for the cloud. HPC-cloud convergence is likely to bridge the gap. In the talk, we will present Streamflow and CAPIO, two development tools for HPC-cloud convergence.},
address = {Catania, Italy},
annote = {keynote},
author = {Marco Aldinucci},
howpublished = {Machine Learning for Astrophysics {(ML4ASTRO)}},
keywords = {deephealth, eupex, across, eupilot},
month = jun,
note = {Keynote talk},
title = {{HPC}-cloud convergence is the missing link between scientific computing and applied{-AI}},
url = {https://datacloud.di.unito.it/index.php/s/2SGswkcip7MoMoH},
year = {2022},
bdsk-url-1 = {https://datacloud.di.unito.it/index.php/s/2SGswkcip7MoMoH}} Da HPC4AI al living lab dello spoke FutureHPC del Centro Nazionale HPC
Marco Aldinucci
across admire eumaster4hpc eupex hpc4ai icsc textarossa
HPC4AI is an open-access laboratory of the University of Turin open to researchers, students and companies that manages a double pair of systems: a production cloud-HPC system and its twin dedicated to development. The cloud-HPC system is implemented thanks to an extended version of the GARR cloud (OpenStack) and the SLURM workload manager. HPC4AI is specifically designed to support system software development and cloud-HPC convergence tools. Among these streamflow (WMS), jupyter-as-a-service (SaaS), portable-secure-tenant (PasS). The experience gained in the design and management of HPC4AI forms the heart of the design of the livinglab of the Turin "FutureHPC" spoke of the National Center "HPC, BigData and Quantum Computing" funded by the PNRR which should be operational from September 2022.
@misc{22:garr,
abstract = {HPC4AI is an open-access laboratory of the University of Turin open to researchers, students and companies that manages a double pair of systems: a production cloud-HPC system and its twin dedicated to development. The cloud-HPC system is implemented thanks to an extended version of the GARR cloud (OpenStack) and the SLURM workload manager. HPC4AI is specifically designed to support system software development and cloud-HPC convergence tools. Among these streamflow (WMS), jupyter-as-a-service (SaaS), portable-secure-tenant (PasS). The experience gained in the design and management of HPC4AI forms the heart of the design of the livinglab of the Turin "FutureHPC" spoke of the National Center "HPC, BigData and Quantum Computing" funded by the PNRR which should be operational from September 2022.},
address = {Palermo, Italy},
annote = {keynote},
author = {Marco Aldinucci},
howpublished = {Condivisioni, Conferenza GARR 2022},
keywords = {hpc4ai, across, eupex, across, admire, textarossa, eumaster4hpc, icsc},
month = may,
note = {Keynote talk},
title = {Da {HPC4AI} al living lab dello spoke {FutureHPC} del Centro Nazionale {HPC}},
url = {https://datacloud.di.unito.it/index.php/s/P3KSroSSmrRxZMc},
year = {2022},
bdsk-url-1 = {https://datacloud.di.unito.it/index.php/s/P3KSroSSmrRxZMc}} OpenDeepHealth: Crafting a Deep Learning Platform as a Service with Kubernetes
Iacopo Colonnelli, Dario Tranchitella
across deephealth hpc4ai jupyter-workflow streamflow
@misc{22:jotb22,
address = {Malaga, Spain},
author = {Iacopo Colonnelli and Dario Tranchitella},
howpublished = {J on The Beach 2022},
keywords = {streamflow, jupyter-workflow, across, deephealth, hpc4ai},
month = apr,
title = {{OpenDeepHealth}: Crafting a Deep Learning Platform as a Service with {K}ubernetes},
url = {https://datacloud.di.unito.it/index.php/s/n6J7STNnwdyqtET},
year = {2022},
bdsk-url-1 = {https://datacloud.di.unito.it/index.php/s/n6J7STNnwdyqtET}} Distributed workflows with Jupyter
Iacopo Colonnelli
across deephealth jupyter-workflow streamflow
@misc{22:jotb22-workshop,
address = {Malaga, Spain},
author = {Iacopo Colonnelli},
howpublished = {J on The Beach 2022},
keywords = {streamflow, jupyter-workflow, deephealth, across},
month = apr,
note = {Workshop},
title = {Distributed workflows with {J}upyter},
url = {https://datacloud.di.unito.it/index.php/s/om89q55S6ePf2Ji},
year = {2022},
bdsk-url-1 = {https://datacloud.di.unito.it/index.php/s/om89q55S6ePf2Ji}} Parallel Computing: a simple introduction, indeed an abstraction
Marco Aldinucci
@misc{22:lincei,
address = {Virtual event},
annote = {invited},
author = {Marco Aldinucci},
howpublished = {I Lincei per la Scuola: scienze per l'innovazione digitale},
month = mar,
note = {Invited talk},
title = {Parallel Computing: a simple introduction, indeed an abstraction},
url = {https://datacloud.di.unito.it/index.php/s/nmaESJrapo2NqHq},
year = {2022},
bdsk-url-1 = {https://datacloud.di.unito.it/index.php/s/nmaESJrapo2NqHq}} 2021
2021
HPC Containers
Iacopo Colonnelli
across
@misc{21:across-containers,
address = {Virtual event},
author = {Iacopo Colonnelli},
howpublished = {ACROSS WP4 meeting},
keywords = {across},
month = jul,
title = {{HPC} Containers},
url = {https://datacloud.di.unito.it/index.php/s/ddf3YBjpm8KBGAF},
year = {2021},
bdsk-url-1 = {https://datacloud.di.unito.it/index.php/s/ddf3YBjpm8KBGAF}} HPC4AI Green Datacenter Design
Marco Aldinucci, Sergio Rabellino
hpc4ai
@misc{21:vertiv,
address = {Milano, Italy},
annote = {invited},
author = {Marco Aldinucci and Sergio Rabellino},
howpublished = {Vertiv keep it running tour},
keywords = {hpc4ai},
month = nov,
note = {Invited talk},
title = {HPC4AI Green Datacenter Design},
url = {https://datacloud.di.unito.it/index.php/s/y6afrJr9w2DTmRN},
year = {2021},
bdsk-url-1 = {https://datacloud.di.unito.it/index.php/s/y6afrJr9w2DTmRN}} HPC4AI: A cloud-HPC ecosystem designed for research and innovation
Marco Aldinucci
hpc4ai
The University of Turin and Polytechnic University of Turin have joined forces to create a federated competence centre on High-Performance Computing (HPC), Artificial Intelligence (AI) and Big Data Analytics (BDA). HPC4AI is designed as a centre capable of collaborating with entrepreneurs to boost their ability to innovate on data-driven technologies and applications. HPC4AI started in 2017 with the construction of four new federated computing laboratories completed at the end of 2020. HPC4AI is organized in two poles: one at the University of Turin (UNITO), which coordinated the design and implementation phase of HPC4AI, and another at the Polytechnic of Turin (POLITO).
@misc{20:dell:hpc4ai,
abstract = {The University of Turin and Polytechnic University of Turin have joined forces to create a federated competence centre on High-Performance Computing (HPC), Artificial Intelligence (AI) and Big Data Analytics (BDA). HPC4AI is designed as a centre capable of collaborating with entrepreneurs to boost their ability to innovate on data-driven technologies and applications. HPC4AI started in 2017 with the construction of four new federated computing laboratories completed at the end of 2020. HPC4AI is organized in two poles: one at the University of Turin (UNITO), which coordinated the design and implementation phase of HPC4AI, and another at the Polytechnic of Turin (POLITO).},
annote = {invited},
author = {Marco Aldinucci},
date-added = {2021-01-21 13:07:40 +0000},
date-modified = {2021-01-21 13:12:29 +0000},
howpublished = {Advanced Computing Workshop 2021: HPC and Beyond},
keywords = {hpc4ai},
month = jan,
note = {Invited talk},
title = {{HPC4AI}: A cloud-{HPC} ecosystem designed for research and innovation},
url = {https://datacloud.di.unito.it/index.php/s/NmM9kB42pJZGMx6},
year = 2021,
bdsk-url-1 = {https://datacloud.di.unito.it/index.php/s/NmM9kB42pJZGMx6}} Lung Nodules Segmentation in CT scans by DeepHealth toolkit
Marco Aldinucci
BDVA
deephealth
@misc{21:icpr:demodeephealth,
address = {Milano. Italy},
annote = {demo},
author = {Marco Aldinucci},
date-added = {2021-01-01 17:49:10 +0100},
date-modified = {2021-01-01 17:49:10 +0100},
howpublished = {25th Intl. Conference on Pattern Recognition},
keywords = {deephealth},
month = jan,
note = {Demo},
publisher = {{BDVA}},
title = {Lung Nodules Segmentation in {CT} scans by DeepHealth toolkit},
url = {https://datacloud.di.unito.it/index.php/s/KYJMcT3pfpat2Hx},
year = 2021,
bdsk-url-1 = {https://datacloud.di.unito.it/index.php/s/KYJMcT3pfpat2Hx}} HPC application cloudification: the streamflow toolkit
Marco Aldinucci
deephealth hpc4ai
@misc{21:parmaditam:hpc4ai,
address = {Virtual event},
annote = {keynote},
author = {Marco Aldinucci},
date-added = {2021-01-01 17:49:10 +0100},
date-modified = {2021-01-01 17:49:10 +0100},
howpublished = {PARMA-DITAM (co-localed with HiPEAC)},
keywords = {hpc4ai, deephealth},
month = jan,
note = {Keynote talk},
title = {{HPC} application cloudification: the streamflow toolkit},
url = {https://datacloud.di.unito.it/index.php/s/HWZijXPqmwfoYCp},
year = 2021,
bdsk-url-1 = {https://datacloud.di.unito.it/index.php/s/HWZijXPqmwfoYCp}} On HPC, AI and their Fatal Attraction
Marco Aldinucci
deephealth hpc4ai
@misc{21:CNR:hpcai,
address = {Virtual event},
annote = {invited},
author = {Marco Aldinucci},
date-added = {2021-01-01 17:49:10 +0100},
date-modified = {2021-01-01 17:49:10 +0100},
howpublished = {CNR IEIIT, Thursday seminars (11 Feb 2021)},
keywords = {hpc4ai, deephealth},
month = feb,
note = {Invited talk},
title = {On {HPC}, {AI} and their Fatal Attraction},
url = {https://datacloud.di.unito.it/index.php/s/pSDxNPncic8gEy8},
year = 2021,
bdsk-url-1 = {https://datacloud.di.unito.it/index.php/s/pSDxNPncic8gEy8}} DeepHealth: Deep Learning ad alte prestazioni per applicazioni in ambito medico
Marco Aldinucci, Marco Beccuti
deephealth hpc4ai
@misc{21:poloict:deephealth,
address = {Torino, Italy},
author = {Marco Aldinucci and Marco Beccuti},
date-added = {2021-08-01 16:19:46 +0200},
date-modified = {2021-08-01 16:20:45 +0200},
howpublished = {Reserach meeting of the PoloICT},
keywords = {deephealth, hpc4ai},
month = apr,
title = {{DeepHealth}: Deep Learning ad alte prestazioni per applicazioni in ambito medico},
url = {https://datacloud.di.unito.it/index.php/s/2F5Net5HdfJTysa},
year = 2021,
bdsk-url-1 = {https://datacloud.di.unito.it/index.php/s/BXdXLzsisQwDLrK}} HPC4AI: Un sistema per la ricerca e l'innovazione dei servizi cloud per l'Intelligenza Artificiale
Marco Aldinucci, Marco Beccuti
hpc4ai
@misc{21:poloict_hpc4ai,
address = {Torino, Italy},
author = {Marco Aldinucci and Marco Beccuti},
date-added = {2021-08-01 16:15:57 +0200},
date-modified = {2021-08-01 16:19:11 +0200},
howpublished = {Reserach meeting of the PoloICT},
keywords = {hpc4ai},
month = apr,
title = {{HPC4AI}: Un sistema per la ricerca e l'innovazione dei servizi cloud per l'Intelligenza Artificiale},
url = {https://datacloud.di.unito.it/index.php/s/BXdXLzsisQwDLrK},
year = 2021,
bdsk-url-1 = {https://datacloud.di.unito.it/index.php/s/BXdXLzsisQwDLrK}} The Universal Cloud-HPC Pipeline for the AI-Assisted Explainable Diagnosis of COVID-19 Pneumonia
Marco Aldinucci, Iacopo Colonnelli
deephealth hpc4ai streamflow
We'll present a methodology to run DNN pipelines on hybrid cloud+HPC infrastructure. We'll also define a "universal pipeline" for medical images. The pipeline can reproduce all state-of-the-art DNNs to diagnose COVID-19 pneumonia, which appeared in the literature during the first Italian lockdown and following months. We can run all of them (across cloud+HPC platforms) and compare their performance in terms of sensitivity and specificity to set a baseline to evaluate future progress in the automated diagnosis of COVID-19. Also, the pipeline makes existing DNNs explainable by way of adversarial training. The pipeline is easily portable and can run across different infrastructures, adapting the performance-urgency trade-off. The methodology builds onto two novel software programs: the streamflow workflow system and the AI-sandbox concept (parallel container with user-space encrypted file system). We reach over 92% accuracy in diagnosing COVID pneumonia.
@misc{21:gtc:clairecovid,
abstract = {We'll present a methodology to run DNN pipelines on hybrid cloud+HPC infrastructure. We'll also define a "universal pipeline" for medical images. The pipeline can reproduce all state-of-the-art DNNs to diagnose COVID-19 pneumonia, which appeared in the literature during the first Italian lockdown and following months. We can run all of them (across cloud+HPC platforms) and compare their performance in terms of sensitivity and specificity to set a baseline to evaluate future progress in the automated diagnosis of COVID-19. Also, the pipeline makes existing DNNs explainable by way of adversarial training. The pipeline is easily portable and can run across different infrastructures, adapting the performance-urgency trade-off. The methodology builds onto two novel software programs: the streamflow workflow system and the AI-sandbox concept (parallel container with user-space encrypted file system). We reach over 92\% accuracy in diagnosing COVID pneumonia.},
address = {Virtual event},
annote = {invited},
author = {Marco Aldinucci and Iacopo Colonnelli},
howpublished = {NVidia GTC'21},
keywords = {streamflow, deephealth, hpc4ai},
month = apr,
note = {Invited talk},
title = {The Universal Cloud-{HPC} Pipeline for the {AI}-Assisted Explainable Diagnosis of {COVID-19} Pneumonia},
url = {https://datacloud.di.unito.it/index.php/s/AkQLbPpEEtDzbbm},
year = 2021,
bdsk-url-1 = {https://datacloud.di.unito.it/index.php/s/AkQLbPpEEtDzbbm}} DeepHealth perspective
Marco Aldinucci
deephealth
TBD
@misc{21:dataweek:deephealth,
abstract = {TBD},
address = {Virtual event},
author = {Marco Aldinucci},
howpublished = {Future challenges in IoT, AI, and convergence of HPC & Cloud & Big Data -- BDVA Data Week},
keywords = {deephealth},
month = may,
title = {DeepHealth perspective},
year = 2021} The Italian research on HPC key technologies across EuroHPC
Marco Aldinucci
ACM Computing Frontiers
across admire eupex eupilot textarossa
High-Performance Computing (HPC) is one of the strategic priorities for research and innovation worldwide due to its relevance for industrial and scientific applications. We envision HPC as composed of three pillars: infrastructures, applications, and key technologies and tools. While infrastructures are by construction centralized in large-scale HPC centers, and applications are generally within the purview of domain-specific organizations, key technologies fall in an intermediate case where coordination is needed, but design and development are often decentralized. A large group of Italian researchers has started a dedicated laboratory within the National Interuniversity Consortium for Informatics (CINI) to address this challenge. The laboratory, albeit young, has managed to succeed in its first attempts to propose a coordinated approach to HPC research within the EuroHPC Joint Undertaking, participating in the calls 2019-20 to five successful proposals for an aggregate total cost of 95M Euro. In this paper, we outline the working group's scope and goals and provide an overview of the five funded projects, which become fully operational in March 2021, and cover a selection of key technologies provided by the working group partners, highlighting their usage development within the projects.
@misc{21:CINI_acm_CF_talk,
abstract = {High-Performance Computing (HPC) is one of the strategic priorities for research and innovation worldwide due to its relevance for industrial and scientific applications. We envision HPC as composed of three pillars: infrastructures, applications, and key technologies and tools. While infrastructures are by construction centralized in large-scale HPC centers, and applications are generally within the purview of domain-specific organizations, key technologies fall in an intermediate case where coordination is needed, but design and development are often decentralized. A large group of Italian researchers has started a dedicated laboratory within the National Interuniversity Consortium for Informatics (CINI) to address this challenge. The laboratory, albeit young, has managed to succeed in its first attempts to propose a coordinated approach to HPC research within the EuroHPC Joint Undertaking, participating in the calls 2019-20 to five successful proposals for an aggregate total cost of 95M Euro. In this paper, we outline the working group's scope and goals and provide an overview of the five funded projects, which become fully operational in March 2021, and cover a selection of key technologies provided by the working group partners, highlighting their usage development within the projects.},
address = {Virtual Conference, Italy},
author = {Marco Aldinucci},
booktitle = {{ACM Computing Frontiers}},
keywords = {across, admire, textarossa, eupex, eupilot},
month = may,
publisher = {{ACM}},
title = {The {Italian} research on {HPC} key technologies across {EuroHPC}},
url = {https://datacloud.di.unito.it/index.php/s/3ZYmDbEm84rbB9k},
year = 2021,
bdsk-url-1 = {https://datacloud.di.unito.it/index.php/s/3ZYmDbEm84rbB9k}} From skeletons to workflows in the cloud-edge era
Marco Aldinucci
across admire deephealth streamflow
Workflows are among the most commonly used tools in a variety of execution environments. Many of them target a specific environment; few of them make it possible to execute an entire workflow in different environments, e.g. Kubernetes and batch clusters. We present a novel approach to workflow execution, called StreamFlow, that complements the workflow graph with the declarative description of potentially complex execution environments and that makes it possible to execute multiple sites not sharing a common data space. StreamFlow supports both task and data parallelism and enables the reproducible and scalable execution of workflows, such as AI pipelines, in hybrid cloud-HPC environments. As a running example, we use the novel “universal COVID-19 pipeline” that explore the whole optimisation space of the training of different DNNs to classify COVID-19 lung lesions.
@misc{21:hlpp:streamflow,
abstract = {Workflows are among the most commonly used tools in a variety of execution environments. Many of them target a specific environment; few of them make it possible to execute an entire workflow in different environments, e.g. Kubernetes and batch clusters. We present a novel approach to workflow execution, called StreamFlow, that complements the workflow graph with the declarative description of potentially complex execution environments and that makes it possible to execute multiple sites not sharing a common data space. StreamFlow supports both task and data parallelism and enables the reproducible and scalable execution of workflows, such as AI pipelines, in hybrid cloud-HPC environments. As a running example, we use the novel ``universal COVID-19 pipeline'' that explore the whole optimisation space of the training of different DNNs to classify COVID-19 lung lesions.},
address = {Virtual event},
annote = {keynote},
author = {Marco Aldinucci},
howpublished = {14th Intl. Symposium on High-Level Programming and Applications (HLPP)},
keywords = {streamflow, deephealth, across, admire},
month = jul,
note = {Keynote talk},
title = {From skeletons to workflows in the cloud-edge era},
url = {https://datacloud.di.unito.it/index.php/s/RyRPjNBse5PKnab},
year = 2021,
bdsk-url-1 = {https://datacloud.di.unito.it/index.php/s/RyRPjNBse5PKnab}} Reproducibility in the AI era
Marco Aldinucci
across admire deephealth
TBD
@misc{21:penta:covid,
abstract = {TBD},
address = {Virtual event},
annote = {invited},
author = {Marco Aldinucci},
howpublished = {Penta Scientific Meeting},
keywords = {deephealth, across, admire},
month = jul,
title = {Reproducibility in the {AI} era},
url = {https://datacloud.di.unito.it/index.php/s/GLpf7kKSJRH733A},
year = 2021,
bdsk-url-1 = {https://datacloud.di.unito.it/index.php/s/GLpf7kKSJRH733A}} The modernization of HPC applications for the cloud era
Marco Aldinucci
across admire deephealth streamflow
Workflows are among the most commonly used tools in a variety of execution environments. Many of them target a specific environment; few of them make it possible to execute an entire workflow in different environments, e.g., clouds, supercomputers, and both of them. We present a novel approach to workflow execution, called StreamFlow, that complements the workflow graph with the declarative description of potentially complex execution environments (such as Kubernetes and SLURM), making it possible to execute onto multiple sites not sharing a common data space. Streamflow clearly distinguishes it from many other workflow management systems because it decouples the data dependencies from the deployment of (containerized) workflow steps. Streamflow also leverages CAPIO (Cross-Application Programmable I/O) to move data from one step to another efficiently. CAPIO captures the POSIX file system and streams it in parallel and in-memory to the workflow's next step, possibly enabling in-transit data filtering.
@misc{21:eni:streamflow,
abstract = {Workflows are among the most commonly used tools in a variety of execution environments. Many of them target a specific environment; few of them make it possible to execute an entire workflow in different environments, e.g., clouds, supercomputers, and both of them. We present a novel approach to workflow execution, called StreamFlow, that complements the workflow graph with the declarative description of potentially complex execution environments (such as Kubernetes and SLURM), making it possible to execute onto multiple sites not sharing a common data space. Streamflow clearly distinguishes it from many other workflow management systems because it decouples the data dependencies from the deployment of (containerized) workflow steps. Streamflow also leverages CAPIO (Cross-Application Programmable I/O) to move data from one step to another efficiently. CAPIO captures the POSIX file system and streams it in parallel and in-memory to the workflow's next step, possibly enabling in-transit data filtering.},
address = {Virtual event},
annote = {keynote},
author = {Marco Aldinucci},
howpublished = {Fifth EAGE Workshop on High Performance Computing for Upstream},
keywords = {streamflow, deephealth, across, admire},
month = sep,
note = {Keynote talk},
title = {The modernization of {HPC} applications for the cloud era},
year = 2021} 2020
2020
StreamFlow: cross breeding cloud with HPC
Iacopo Colonnelli
streamflow
Workflows are among the most commonly used tools in a variety of execution environments. Many of them target a specific environment; few of them make it possible to execute an entire workflow in different environments, e.g. Kubernetes and batch clusters. We present a novel approach to workflow execution, called StreamFlow, that complements the workflow graph with the declarative description of potentially complex execution environments, and that makes it possible the execution onto multiple sites not sharing a common data space. StreamFlow is then exemplified on a novel bioinformatics pipeline for single-cell transcriptomic data analysis workflow.
@misc{20:HPCEuropa3TAM,
abstract = {Workflows are among the most commonly used tools in a variety of execution environments. Many of them target a specific environment; few of them make it possible to execute an entire workflow in different environments, e.g. Kubernetes and batch clusters. We present a novel approach to workflow execution, called StreamFlow, that complements the workflow graph with the declarative description of potentially complex execution environments, and that makes it possible the execution onto multiple sites not sharing a common data space. StreamFlow is then exemplified on a novel bioinformatics pipeline for single-cell transcriptomic data analysis workflow.},
address = {Virtual event},
annote = {invited},
author = {Iacopo Colonnelli},
bdsk-utl-2 = {https://drive.google.com/file/d/1aTVhlsrS7R2FzEWRTTtm4Mzr7O8R4Jq1/view},
howpublished = {HPC-Europa3 2nd Transnational Access Meeting (TAM)},
keywords = {streamflow},
month = oct,
note = {Invited talk},
title = {{StreamFlow}: cross breeding cloud with {HPC}},
url = {https://datacloud.di.unito.it/index.php/s/qPHHrSNxk8QXJDw},
year = {2020},
bdsk-url-1 = {https://datacloud.di.unito.it/index.php/s/qPHHrSNxk8QXJDw}} JupyterFlow: Jupyter Notebooks su larga scala
Iacopo Colonnelli, Sergio Rabellino
deephealth hpc4ai jupyter-workflow
I Jupyter Notebook sono largamente utilizzati sia in ambito industriale che accademico come strumento di didattica, prototipazione e analisi esplorative. Purtroppo il sistema runtime standard di Jupyter non è abbastanza potente per sostenere un carichi di lavoro reali e spesso l'unica soluzione è quella di riscrivere il codice da zero in una tecnologia con supporto HPC. Intrgrando lo stack Jupyter con StreamFlow (https://streamflow.di.unito.it/) è possibile creare i Notebook tramite un'interfaccia web su cloud ed eseguirli in maniera trasparente in remoto su una VM con GPU o su nodi HPC.
@misc{20:GarrWorkshop,
abstract = {I Jupyter Notebook sono largamente utilizzati sia in ambito industriale che accademico come strumento di didattica, prototipazione e analisi esplorative. Purtroppo il sistema runtime standard di Jupyter non \`{e} abbastanza potente per sostenere un carichi di lavoro reali e spesso l'unica soluzione \`{e} quella di riscrivere il codice da zero in una tecnologia con supporto HPC. Intrgrando lo stack Jupyter con StreamFlow (https://streamflow.di.unito.it/) \`{e} possibile creare i Notebook tramite un'interfaccia web su cloud ed eseguirli in maniera trasparente in remoto su una VM con GPU o su nodi HPC.},
address = {Virtual event},
author = {Iacopo Colonnelli and Sergio Rabellino},
howpublished = {Workshop GARR 2020},
keywords = {jupyter-workflow, hpc4ai, deephealth},
month = nov,
title = {{JupyterFlow}: {J}upyter {N}otebooks su larga scala},
url = {https://datacloud.di.unito.it/index.php/s/ASPEmyXAj5QscgC},
year = {2020},
bdsk-url-1 = {https://datacloud.di.unito.it/index.php/s/ASPEmyXAj5QscgC},
bdsk-url-2 = {https://www.eventi.garr.it/it/ws20/programma/speaker/680-iacopo-colonnelli}} Mnemocomputing
Marco Aldinucci
@misc{20:hlpgpu:mnemocomputing,
address = {Valencia, Spain},
annote = {keynote},
author = {Marco Aldinucci},
date-added = {2021-01-01 17:49:10 +0100},
date-modified = {2021-01-01 17:49:10 +0100},
howpublished = {HLPGPU 2019 (Satellite workshop of HiPEAC 2019)},
month = jan,
note = {Keynote talk},
title = {Mnemocomputing},
url = {https://datacloud.di.unito.it/index.php/s/8pAGGsw4nEtrtNj},
year = 2020,
bdsk-url-1 = {https://datacloud.di.unito.it/index.php/s/8pAGGsw4nEtrtNj}} Machine Learning: the treacherous journey from data to knowledge (with examples from HPC4AIUNITO platform)
Marco Aldinucci
deephealth hpc4ai
@misc{20:chem:HPCAI,
address = {Torino, Italy},
annote = {invited},
author = {Marco Aldinucci},
date-added = {2021-01-01 17:49:10 +0100},
date-modified = {2021-01-01 17:49:10 +0100},
howpublished = {Machine Learning Meets Chemistry \@ the Department of Chemistry, University of Torino},
keywords = {hpc4ai, deephealth},
month = feb,
note = {Invited talk},
title = {Machine Learning: the treacherous journey from data to knowledge (with examples from {HPC4AI\@UNITO} platform)},
url = {https://datacloud.di.unito.it/index.php/s/ffyZYYqNQpkza4F},
year = 2020,
bdsk-url-1 = {https://datacloud.di.unito.it/index.php/s/ffyZYYqNQpkza4F}} HPC4AI: From Enabling Platforms to Technology Sovereignty to Innovation
Marco Aldinucci
hpc4ai
@misc{20:elixir:hpc4ai,
address = {Torino, Italy},
annote = {invited},
author = {Marco Aldinucci},
date-added = {2021-01-01 17:49:10 +0100},
date-modified = {2021-01-01 17:49:10 +0100},
howpublished = {Elixir-Italia meeting},
keywords = {hpc4ai},
month = feb,
note = {Invited talk},
title = {{HPC4AI}: From Enabling Platforms to Technology Sovereignty to Innovation},
url = {https://datacloud.di.unito.it/index.php/s/g6LZiErXH4PPRCj},
year = 2020,
bdsk-url-1 = {https://datacloud.di.unito.it/index.php/s/g6LZiErXH4PPRCj}} The DeepHealth project
Marco Aldinucci
BDVA
deephealth
@misc{20:bdva:deephealth,
annote = {invited},
author = {Marco Aldinucci},
date-added = {2021-01-01 17:49:10 +0100},
date-modified = {2021-02-14 18:48:35 +0100},
howpublished = {HPC, Big Data, IoT and AI future industry-driven collaborative strategic topics virtual workshop --- HPC / HPDA spectrum},
keywords = {deephealth},
month = jul,
note = {Invited talk},
publisher = {BDVA},
title = {The DeepHealth project},
url = {https://datacloud.di.unito.it/index.php/s/ortwLJHS2q96irb},
year = 2020,
bdsk-url-1 = {https://datacloud.di.unito.it/index.php/s/ortwLJHS2q96irb}} Building avenues for AI-assisted diagnosis over the bridge from HPC to AI
Marco Aldinucci
deephealth
@misc{20:claire:taskforce,
address = {Torino, Italy},
annote = {invited},
author = {Marco Aldinucci},
date-added = {2021-01-01 17:49:10 +0100},
date-modified = {2021-01-01 17:49:10 +0100},
howpublished = {CLAIRE COVID webinar},
keywords = {deephealth},
month = jul,
note = {Invited talk},
title = {Building avenues for AI-assisted diagnosis over the bridge from {HPC} to {AI}},
url = {https://datacloud.di.unito.it/index.php/s/RqpNCHyyL6wc5ds},
year = 2020,
bdsk-url-1 = {https://datacloud.di.unito.it/index.php/s/RqpNCHyyL6wc5ds}} 2019
2019
Deep Learning at Scale
Iacopo Colonnelli
IEEE
This work presents a novel approach to distributed training of deep neural networks (DNNs) that aims to overcome the issues related to mainstream approaches to data parallel training. Established techniques for data parallel training are discussed from both a parallel computing and deep learning perspective, then a different approach is presented that is meant to allow DNN training to scale while retaining good convergence properties. Moreover, an experimental implementation is presented as well as some preliminary results.
@misc{19:PDPNNT,
abstract = {This work presents a novel approach to distributed training of deep neural networks (DNNs) that aims to overcome the issues related to mainstream approaches to data parallel training. Established techniques for data parallel training are discussed from both a parallel computing and deep learning perspective, then a different approach is presented that is meant to allow DNN training to scale while retaining good convergence properties. Moreover, an experimental implementation is presented as well as some preliminary results.},
address = {Pavia, Italy},
author = {Iacopo Colonnelli},
howpublished = {27th Euromicro International Conference on Parallel, Distributed and Network-based Processing (PDP 2019)},
month = feb,
publisher = {{IEEE}},
title = {Deep Learning at Scale},
url = {https://datacloud.di.unito.it/index.php/s/nRW9M69C3AtpDoM},
year = {2019},
bdsk-url-1 = {https://datacloud.di.unito.it/index.php/s/nRW9M69C3AtpDoM}} Accelerating spectral graph analysis through wavefronts of linear algebra operations
Iacopo Colonnelli
IEEE
The wavefront pattern captures the unfolding of a parallel computation in which data elements are laid out as a logical multidimensional grid and the dependency graph favours a diagonal sweep across the grid. In the emerging area of spectral graph analysis, the computing often consists in a wavefront running over a tiled matrix, involving expensive linear algebra kernels. While these applications might benefit from parallel heterogeneous platforms (multi-core with GPUs),programming wavefront applications directly with high-performance linear algebra libraries yields code that is complex to write and optimize for the specific application. We advocate a methodology based on two abstractions (linear algebra and parallel pattern-based run-time), that allows to develop portable, self-configuring, and easy-to-profile code on hybrid platforms.
@misc{19:PDPArmadillo,
abstract = {The wavefront pattern captures the unfolding of a parallel computation in which data elements are laid out as a logical multidimensional grid and the dependency graph favours a diagonal sweep across the grid. In the emerging area of spectral graph analysis, the computing often consists in a wavefront running over a tiled matrix, involving expensive linear algebra kernels. While these applications might benefit from parallel heterogeneous platforms (multi-core with GPUs),programming wavefront applications directly with high-performance linear algebra libraries yields code that is complex to write and optimize for the specific application. We advocate a methodology based on two abstractions (linear algebra and parallel pattern-based run-time), that allows to develop portable, self-configuring, and easy-to-profile code on hybrid platforms.},
address = {Pavia, Italy},
author = {Iacopo Colonnelli},
howpublished = {27th Euromicro International Conference on Parallel, Distributed and Network-based Processing (PDP 2019)},
month = feb,
publisher = {{IEEE}},
title = {Accelerating spectral graph analysis through wavefronts of linear algebra operations},
url = {https://datacloud.di.unito.it/index.php/s/zK4eSzdsdB8CfQX},
year = {2019},
bdsk-url-1 = {https://datacloud.di.unito.it/index.php/s/zK4eSzdsdB8CfQX}} StreamFlow: un approccio dichiarativo a workflow e pipeline di micro-servizi
Iacopo Colonnelli
streamflow
Negli ultimi anni, gli approcci orientati ai container si sono dimostrati particolarmente efficaci nel garantire portabilità e riproducibilità dei workflow scientifici. Tuttavia, con il continuo aumento del volume di dati a disposizione e la crescente complessità delle procedure di analisi in ogni campo della ricerca, anche i requisiti di performance e riusabilità si fanno via via sempre più essenziali. L'obiettivo principale di StreamFlow è quello di fornire un nuovo paradigma, totalmente dichiarativo, per la descrizione e l'accelerazione di workflow scientifici in ambienti distribuiti. La peculiarità di StreamFlow risiede nel fatto che l'ambiente di esecuzione è interamente descritto in termini di servizi (container), connessioni tra essi e fattori di replica. Inoltre, ogni task del workflow è esplicitamente mappato sulla tipologia di servizio richiesta. Questo permette un maggior controllo sull'utilizzo delle risorse e politiche di scheduling più precise, a vantaggio delle performance. I principali vantaggi di un approccio dichiarativo sono invece la più facile comprensione ed estensione dei modelli esistenti, a vantaggio della riusabilitià.
@misc{19:GarrWorkshop,
abstract = {Negli ultimi anni, gli approcci orientati ai container si sono dimostrati particolarmente efficaci nel garantire portabilit\`{a} e riproducibilit\`{a} dei workflow scientifici. Tuttavia, con il continuo aumento del volume di dati a disposizione e la crescente complessit\`{a} delle procedure di analisi in ogni campo della ricerca, anche i requisiti di performance e riusabilit\`{a} si fanno via via sempre pi\`{u} essenziali. L'obiettivo principale di StreamFlow \`{e} quello di fornire un nuovo paradigma, totalmente dichiarativo, per la descrizione e l'accelerazione di workflow scientifici in ambienti distribuiti. La peculiarit\`{a} di StreamFlow risiede nel fatto che l'ambiente di esecuzione \`{e} interamente descritto in termini di servizi (container), connessioni tra essi e fattori di replica. Inoltre, ogni task del workflow \`{e} esplicitamente mappato sulla tipologia di servizio richiesta. Questo permette un maggior controllo sull'utilizzo delle risorse e politiche di scheduling pi\`{u} precise, a vantaggio delle performance. I principali vantaggi di un approccio dichiarativo sono invece la pi\`{u} facile comprensione ed estensione dei modelli esistenti, a vantaggio della riusabiliti\`{a}.},
address = {Roma, Italy},
author = {Iacopo Colonnelli},
howpublished = {Workshop GARR 2019},
keywords = {streamflow},
month = oct,
title = {{StreamFlow}: un approccio dichiarativo a workflow e pipeline di micro-servizi},
url = {https://datacloud.di.unito.it/index.php/s/kZqyiQnBEQNdXJe},
year = {2019},
bdsk-url-1 = {https://datacloud.di.unito.it/index.php/s/kZqyiQnBEQNdXJe},
bdsk-url-2 = {https://www.eventi.garr.it/it/ws19/programma/speaker/580-iacopo-colonnelli}} The evolution of high-performance systems: from HPC to Big Data to Deep Learning
Marco Aldinucci
@misc{19:SmartData:fromHPCtoDL,
address = {Torino, Italy},
annote = {invited},
author = {Marco Aldinucci},
date-added = {2021-03-17 15:00:00 +0100},
date-modified = {2021-03-17 15:00:00 +0100},
howpublished = {4th Open SmartData\@PoliTO Workshop},
month = feb,
note = {Invited talk},
title = {The evolution of high-performance systems: from HPC to Big Data to Deep Learning},
url = {https://datacloud.di.unito.it/index.php/s/cZitMsp5GPJ2Qzf},
year = 2019,
bdsk-url-1 = {https://datacloud.di.unito.it/index.php/s/cZitMsp5GPJ2Qzf}} High-Performance Computing
Marco Aldinucci
@misc{19:iCom:HPC,
address = {Roma, Italy},
annote = {invited},
author = {Marco Aldinucci},
date-added = {2021-03-17 15:00:00 +0100},
date-modified = {2021-03-17 15:00:00 +0100},
howpublished = {Istituto per la Competitivit{\`a}},
month = apr,
note = {Invited talk},
title = {High-Performance Computing},
url = {https://www.i-com.it/2019/02/02/supercomputer-hpe-computing/},
year = 2019,
bdsk-url-1 = {}} HPC4AI, an on-demand federated platform endeavour
Marco Aldinucci
hpc4ai
@misc{19:SR:hpc4ai,
address = {Milano, Italy},
annote = {invited},
author = {Marco Aldinucci},
date-added = {2021-01-01 17:49:10 +0100},
date-modified = {2021-01-01 17:49:10 +0100},
howpublished = {Ospedale San Raffaele},
keywords = {hpc4ai},
month = may,
note = {Invited talk},
title = {{HPC4AI}, an on-demand federated platform endeavour},
url = {https://datacloud.di.unito.it/index.php/s/P7fxgExkJDAFQbm},
year = 2019,
bdsk-url-1 = {https://datacloud.di.unito.it/index.php/s/P7fxgExkJDAFQbm}} L'infrastruttura necessaria per creare interoperabilità tra pubbliche amministrazioni
Marco Aldinucci
@misc{19:conference:PublicBD,
address = {Turin, Italy},
annote = {invited},
author = {Marco Aldinucci},
date-added = {2021-03-17 14:00:00 +0100},
date-modified = {2021-03-17 14:00:00 +0100},
howpublished = {Convegno ``L'amministrazione pubblica con i Big data''},
month = may,
note = {Invited talk},
title = {L'infrastruttura necessaria per creare interoperabilit{\`a} tra pubbliche amministrazioni},
url = {https://datacloud.di.unito.it/index.php/s/KPebr76m88jQATi},
year = 2019,
bdsk-url-1 = {https://datacloud.di.unito.it/index.php/s/KPebr76m88jQATi}} Towards DeepLearning at Scale
Marco Aldinucci
@misc{19:SPPEXA:DLscale,
address = {Dresden, Germany},
author = {Marco Aldinucci},
date-added = {2021-03-17 14:00:00 +0100},
date-modified = {2021-03-17 14:00:00 +0100},
howpublished = {SPPEXA Final Symposium},
month = oct,
note = {Invited talk},
title = {Towards DeepLearning at Scale},
year = 2019,
bdsk-url-1 = {}} The Joint Undertaking EuroHPC
Marco Aldinucci
@misc{19:APRE:EuroHPC,
address = {Roma, Italy},
author = {Marco Aldinucci},
date-added = {2021-03-17 14:00:00 +0100},
date-modified = {2021-03-17 14:00:00 +0100},
howpublished = {Giornata Nazionale di Lancio dei bandi Information and Communication Technologies in Horizon 2020},
month = oct,
note = {Invited talk},
title = {The Joint Undertaking EuroHPC},
year = 2019,
bdsk-url-1 = {}} How Artificial Intelligence is shaping the Future of Health
Marco Aldinucci
@misc{19:ICAIH:AIhealth,
address = {Milano, Italy},
annote = {invited},
author = {Marco Aldinucci},
date-added = {2021-03-16 15:00:00 +0100},
date-modified = {2021-03-16 15:00:00 +0100},
howpublished = {1st Industrial Conference on Artificial Intelligence and health},
month = nov,
note = {Invited talk},
title = {How Artificial Intelligence is shaping the Future of Health},
url = {https://datacloud.di.unito.it/index.php/s/FPDCjcWwprAa7D7},
year = 2019,
bdsk-url-1 = {https://datacloud.di.unito.it/index.php/s/FPDCjcWwprAa7D7}} 2018
2018
Toward Near-Data Processing service computing
Marco Aldinucci
IEEE
@misc{18:PDP:NDP,
address = {Cambridge, UK},
annote = {http://www.pdp2018.org/invited.html, invited, keynote},
author = {Marco Aldinucci},
date-added = {2021-01-01 18:02:39 +0100},
date-modified = {2021-01-01 18:54:26 +0100},
howpublished = {Parallel, Distributed, and Network-Based Processing (PDP)},
month = mar,
note = {Keynote talk},
publisher = {{IEEE}},
title = {Toward Near-Data Processing service computing},
url = {https://datacloud.di.unito.it/index.php/s/ePDZ44KSsgXSGHw},
year = 2018,
bdsk-url-1 = {https://datacloud.di.unito.it/index.php/s/ePDZ44KSsgXSGHw}} From mutexes to lock-free to atomic-less and back to (un-contended) mutex: A story in which energy efficiency is unexpectedly constantly increasing
Marco Aldinucci
@misc{18:ARM:energyEfficiency,
address = {Cambridge, UK},
author = {Marco Aldinucci},
date-added = {2021-03-18 15:00:00 +0100},
date-modified = {2021-03-18 15:00:00 +0100},
howpublished = {Arm Research ltd},
month = mar,
note = {Invited talk},
title = {From mutexes to lock-free to atomic-less and back to (un-contended) mutex: A story in which energy efficiency is unexpectedly constantly increasing},
year = 2018,
bdsk-url-1 = {}} Designing a heterogeneous federated data center for research
Marco Aldinucci
@misc{18:DISIA:federated,
address = {Firenze, Italy},
author = {Marco Aldinucci},
date-added = {2021-03-18 15:00:00 +0100},
date-modified = {2021-03-18 15:00:00 +0100},
howpublished = {DiSIA, University of Florence},
month = mar,
note = {Invited talk},
title = {Designing a heterogeneous federated data center for research},
year = 2018,
bdsk-url-1 = {}} HPC4AI, an AI-on-demand federated platform endeavour
Marco Aldinucci
@misc{18:ACM:HPC4AI,
address = {Ischia, Italy},
annote = {invited},
author = {Marco Aldinucci},
date-added = {2021-03-18 12:00:00 +0100},
date-modified = {2021-03-18 12:00:00 +0100},
howpublished = {ACM Computing Frontiers},
month = may,
note = {Invited talk},
title = {HPC4AI, an AI-on-demand federated platform endeavour},
url = {https://datacloud.di.unito.it/index.php/s/NWbZ2FYpeYR47T6},
year = 2018,
bdsk-url-1 = {https://datacloud.di.unito.it/index.php/s/NWbZ2FYpeYR47T6}} Le piattaforme AI-on-demand come fattore di l'innovazione nelle PMI
Marco Aldinucci
@misc{18:CITYLAB:AIonDemand,
address = {Torino, Italy},
annote = {invited},
author = {Marco Aldinucci},
date-added = {2021-03-18 12:00:00 +0100},
date-modified = {2021-03-18 12:00:00 +0100},
howpublished = {CITYLAB Ecosystem: i dati per le aziende, le citt{\`a} e le universit{\`a}, Nuvola Lavazza Torino},
month = jun,
note = {Invited talk},
title = {Le piattaforme AI-on-demand come fattore di l'innovazione nelle PMI},
url = {https://datacloud.di.unito.it/index.php/s/Twy9zwbJB9KnBE2},
year = 2018,
bdsk-url-1 = {https://datacloud.di.unito.it/index.php/s/Twy9zwbJB9KnBE2}} L'evoluzione delle piattaforme e dei sistemi ad alte prestazioni: da HPC ai Big Data al Deep Learning
Marco Aldinucci
@misc{18:ChimicaPP:fromHPCtoDL,
address = {Torino, Italy},
annote = {invited},
author = {Marco Aldinucci},
date-added = {2021-03-18 12:00:00 +0100},
date-modified = {2021-03-18 12:00:00 +0100},
howpublished = {Chimica passione periodica: Big Data: Modelli predittivi, simulazione, analisi, Dipartimento di Chimica, Universit{\`a} di Torino},
month = nov,
note = {Invited talk},
title = {L'evoluzione delle piattaforme e dei sistemi ad alte prestazioni: da HPC ai Big Data al Deep Learning},
url = {https://datacloud.di.unito.it/index.php/s/sLxm6G2Ma5jj9BP},
year = 2018,
bdsk-url-1 = {https://datacloud.di.unito.it/index.php/s/sLxm6G2Ma5jj9BP}} 2017
2017
Streaming in the PGAS Era
Marco Aldinucci
@misc{17:autodasp:17,
address = {Santiago de Compostela, Spain},
annote = {http://calvados.di.unipi.it/auto-dasp-17/index.php/program/, keynote},
author = {Marco Aldinucci},
date-added = {2021-01-01 19:04:30 +0100},
date-modified = {2021-01-01 19:07:40 +0100},
howpublished = {International Workshop on Autonomic Solutions for Parallel and Distributed Data Stream Processing (Auto-DaSP 2017), Workshop of EuroPar 2017},
month = aug,
note = {Keynote talk},
title = {Streaming in the {PGAS} Era},
url = {https://datacloud.di.unito.it/index.php/s/H6ZRoK6gLZDQFta},
year = 2017,
bdsk-url-1 = {https://datacloud.di.unito.it/index.php/s/H6ZRoK6gLZDQFta}} Partitioned Global Address Space in the mainstream of C++ programming
Marco Aldinucci
@misc{20:parco:PGAS,
address = {Bologna, Italy},
annote = {http://www.parco.org/keynotes.html, keynote},
author = {Marco Aldinucci},
date-added = {2021-01-01 19:00:41 +0100},
date-modified = {2021-01-01 19:04:21 +0100},
howpublished = {International Parallel Computing (ParCo)},
month = sep,
note = {Keynote talk},
title = {Partitioned Global Address Space in the mainstream of {C++} programming},
url = {https://datacloud.di.unito.it/index.php/s/dTAm7oAfHRTY7BC},
year = 2017,
bdsk-url-1 = {https://datacloud.di.unito.it/index.php/s/dTAm7oAfHRTY7BC}} OCCAM: Heterogeneous platforms mixed blessing of code optimisation
Marco Aldinucci
@misc{17:INFN:occam,
address = {Torino, Italy},
annote = {invited},
author = {Marco Aldinucci},
date-added = {2021-03-18 16:00:00 +0100},
date-modified = {2021-03-18 16:00:00 +0100},
howpublished = {Riunione della Commissione Calcolo e Reti dell'INFN},
month = sep,
note = {Invited talk},
title = {OCCAM: Heterogeneous platforms mixed blessing of code optimisation},
url = {https://agenda.infn.it/login/?next=%2Fevent%2F12983%2Ftimetable%2F%3Fview%3Dstandard},
year = 2017,
bdsk-url-1 = {https://agenda.infn.it/login/?next=%2Fevent%2F12983%2Ftimetable%2F%3Fview%3Dstandard}} 2016
2016
FastDataUNITO opening
Marco Aldinucci
@misc{16:UNITO:fastdata,
address = {Torino, Italy},
annote = {invited},
author = {Marco Aldinucci},
date-added = {2021-03-18 16:00:00 +0100},
date-modified = {2021-03-18 16:00:00 +0100},
month = mar,
note = {Invited talk},
title = {FastData\@UNITO opening},
year = 2016,
bdsk-url-1 = {}} Composition and Compartmentalisation As Enabling Features for Data-Centric, Extreme Scale Applications: An MPI-X Approach
Marco Aldinucci
@misc{16:SIAM:composition,
address = {Paris, France},
author = {Marco Aldinucci},
date-added = {2021-03-18 16:00:00 +0100},
date-modified = {2021-03-18 16:00:00 +0100},
howpublished = {SIAM Conference on Parallel Processing for Scientific Computing},
month = apr,
title = {Composition and Compartmentalisation As Enabling Features for Data-Centric, Extreme Scale Applications: An MPI-X Approach},
url = {https://datacloud.di.unito.it/index.php/s/cWne6zEWypepo7H},
year = 2016,
bdsk-url-1 = {https://datacloud.di.unito.it/index.php/s/cWne6zEWypepo7H}} I supercomputer venuti dal futuro: Exascale Computing
Marco Aldinucci
@misc{16:PintOfScience:exascale,
address = {Torino, Italy},
author = {Marco Aldinucci},
date-added = {2021-03-18 16:00:00 +0100},
date-modified = {2021-03-18 16:00:00 +0100},
howpublished = {International Pint of Science, Officine ferroviarie},
month = may,
note = {Invited talk},
title = {I supercomputer venuti dal futuro: Exascale Computing},
year = 2016,
bdsk-url-1 = {}} Atomic operations considered hARMful
Marco Aldinucci
@misc{16:ARM:atomic,
address = {Cambridge, UK},
author = {Marco Aldinucci},
date-added = {2021-03-18 16:00:00 +0100},
date-modified = {2021-03-18 16:00:00 +0100},
howpublished = {ARM Research Summary},
month = sep,
title = {Atomic operations considered hARMful},
url = {https://datacloud.di.unito.it/index.php/s/mBoWAmMzj8nYrbe},
year = 2016,
bdsk-url-1 = {https://datacloud.di.unito.it/index.php/s/mBoWAmMzj8nYrbe}} HPC come piattaforma abilitante: rischi e opportunità
Marco Aldinucci
@misc{16:CARLOALBERTO:dataScience,
address = {Torino, Italy},
annote = {invited},
author = {Marco Aldinucci},
date-added = {2021-03-18 16:00:00 +0100},
date-modified = {2021-03-18 16:00:00 +0100},
howpublished = {Workshop of the Competence Centre on Scientific Computing of University of Torino},
month = oct,
note = {Invited talk},
title = {HPC come piattaforma abilitante: rischi e opportunit{\`a}},
url = {https://datacloud.di.unito.it/index.php/s/6qgoQoSXaZpMQq4},
year = 2016,
bdsk-url-1 = {https://datacloud.di.unito.it/index.php/s/6qgoQoSXaZpMQq4}} 2014
2014
An Overview of FastFlow: Combining Pattern-Level Abstraction and Efficiency in GPGPUs
Marco Aldinucci
@misc{14:GTC:overview,
address = {San Jose, CA, USA},
author = {Marco Aldinucci},
date-added = {2021-03-18 16:00:00 +0100},
date-modified = {2021-03-18 16:00:00 +0100},
howpublished = {GPU Technology Conference (GTC 2014)},
month = mar,
title = {An Overview of FastFlow: Combining Pattern-Level Abstraction and Efficiency in GPGPUs},
url = {https://datacloud.di.unito.it/index.php/s/sRYnknEHn3Qkj5b},
year = 2014,
bdsk-url-1 = {https://datacloud.di.unito.it/index.php/s/sRYnknEHn3Qkj5b}} FastFlow: Combining Pattern-Level Abstraction and Efficiency in GPGPUs
Marco Aldinucci
@misc{14:GTC:fastflow,
address = {San Jose, CA, USA},
author = {Marco Aldinucci},
date-added = {2021-03-18 16:00:00 +0100},
date-modified = {2021-03-18 16:00:00 +0100},
howpublished = {GPU Technology Conference (GTC 2014)},
month = mar,
title = {FastFlow: Combining Pattern-Level Abstraction and Efficiency in GPGPUs},
url = {https://datacloud.di.unito.it/index.php/s/ibCGJc7tJyASMbN},
year = 2014,
bdsk-url-1 = {https://datacloud.di.unito.it/index.php/s/ibCGJc7tJyASMbN}} FastFlow: high-level programming patterns with non-blocking lock-free run-time support
Marco Aldinucci
@misc{14:dublin:ff,
address = {Dublin, Ireland},
annote = {keynote},
author = {Marco Aldinucci},
date-added = {2021-01-01 17:49:10 +0100},
date-modified = {2021-02-01 11:32:48 +0100},
howpublished = {Intl. Summer School in Parallel Patterns},
month = jun,
note = {Keynote talk},
title = {FastFlow: high-level programming patterns with non-blocking lock-free run-time support},
url = {https://datacloud.di.unito.it/index.php/s/TJP46RfwX2NGPTS},
year = 2014,
bdsk-url-1 = {https://datacloud.di.unito.it/index.php/s/TJP46RfwX2NGPTS}} Parallel patterns, data-centric concurrency, and heterogeneous computing
Marco Aldinucci
@misc{14:hpcc:datacentric,
address = {Paris, France},
annote = {keynote},
author = {Marco Aldinucci},
date-added = {2021-01-01 17:49:10 +0100},
date-modified = {2021-01-01 17:49:10 +0100},
howpublished = {6th IEEE Intl. Conference on High-Performance Computing and Communications (HPCC)},
month = aug,
note = {Keynote talk},
title = {Parallel patterns, data-centric concurrency, and heterogeneous computing},
url = {https://datacloud.di.unito.it/index.php/s/mRSTsB6qjKMYiEN},
year = 2014,
bdsk-url-1 = {https://datacloud.di.unito.it/index.php/s/mRSTsB6qjKMYiEN}} 2013
2013
Turning Big data into knowledge: systems biology droplets in the cloud
Marco Aldinucci
@misc{13:RAF:cloud,
address = {Madrid, Spain},
annote = {invited},
author = {Marco Aldinucci},
date-added = {2021-03-18 17:00:00 +0100},
date-modified = {2021-03-18 17:00:00 +0100},
howpublished = {Cloud and Science seminar at ``the Ramon Areces Foundation''},
month = mar,
note = {invited talk},
title = {Turning Big data into knowledge: systems biology droplets in the cloud},
url = {https://datacloud.di.unito.it/index.php/s/oBPqA5DeyCbj5do},
year = 2013,
bdsk-url-1 = {https://datacloud.di.unito.it/index.php/s/oBPqA5DeyCbj5do}} 2012
2012
Pattern-based Parallel Edge Preserving Algorithm for Salt-and-Pepper Image Denoising
Marco Aldinucci
@misc{12:HPCAC:denoising,
address = {Lugano, Switzerland},
annote = {invited},
author = {Marco Aldinucci},
date-added = {2021-03-18 17:00:00 +0100},
date-modified = {2021-03-18 17:00:00 +0100},
howpublished = {HPC Advisory Council Switzerland Conference},
month = mar,
note = {invited talk},
title = {Pattern-based Parallel Edge Preserving Algorithm for Salt-and-Pepper Image Denoising},
url = {https://datacloud.di.unito.it/index.php/s/BCm7w9S86yDNJb2},
year = 2012,
bdsk-url-1 = {https://datacloud.di.unito.it/index.php/s/BCm7w9S86yDNJb2}} Turning Big data into knowledge: Techniques and Tools for Parallel Computing on Online Data Streams in Systems Biology and Epidemiology
Marco Aldinucci
@misc{12:bigdata:ff,
address = {Rhodes, Greece},
annote = {keynote},
author = {Marco Aldinucci},
date-added = {2021-01-01 17:49:10 +0100},
date-modified = {2021-02-01 11:21:34 +0100},
howpublished = {Workshop in big data management},
month = aug,
title = {Turning Big data into knowledge: Techniques and Tools for Parallel Computing on Online Data Streams in Systems Biology and Epidemiology},
url = {https://datacloud.di.unito.it/index.php/s/yg7i63MQdr7gZt7},
year = 2012,
bdsk-url-1 = {https://datacloud.di.unito.it/index.php/s/yg7i63MQdr7gZt7}} A Parallel Edge Preserving Algorithm for Salt and Pepper Image Denoising
Marco Aldinucci
@misc{12:IEEE:denoising,
address = {Istanbul, Turkey},
author = {Marco Aldinucci},
date-added = {2021-03-18 17:00:00 +0100},
date-modified = {2021-03-18 17:00:00 +0100},
howpublished = {IEEE Intl. Conference on Image Processing Theory, Tools and Applications (IPTA)},
month = oct,
title = {A Parallel Edge Preserving Algorithm for Salt and Pepper Image Denoising},
url = {https://datacloud.di.unito.it/index.php/s/JQe8RDp9jp4iymF},
year = 2012,
bdsk-url-1 = {https://datacloud.di.unito.it/index.php/s/JQe8RDp9jp4iymF}} 2011
2011
On Designing Multicore-Aware Simulators for Biological Systems
Marco Aldinucci
@misc{11:IEEE:simulator,
address = {Ayia Napa, Ciprus},
author = {Marco Aldinucci},
date-added = {2021-03-18 18:00:00 +0100},
date-modified = {2021-03-18 18:00:00 +0100},
howpublished = {IEEE Euromicro PDP 2011: Parallel Distributed and network-based Processing},
month = feb,
title = {On Designing Multicore-Aware Simulators for Biological Systems},
url = {https://datacloud.di.unito.it/index.php/s/Anb4Gdj8p9yq4C2},
year = 2011,
bdsk-url-1 = {https://datacloud.di.unito.it/index.php/s/Anb4Gdj8p9yq4C2}} High-level parallel programming: (few) ideas for challenges in formal methods
Marco Aldinucci
@misc{11:COST:parallel,
address = {Limerick, Republic of Ireland},
annote = {invited},
author = {Marco Aldinucci},
date-added = {2021-03-18 17:00:00 +0100},
date-modified = {2021-03-18 17:00:00 +0100},
howpublished = {COST Action IC701 workshop},
month = jun,
note = {invited talk},
title = {High-level parallel programming: (few) ideas for challenges in formal methods},
url = {https://datacloud.di.unito.it/index.php/s/XfikGmd2e6pwfaM},
year = 2011,
bdsk-url-1 = {https://datacloud.di.unito.it/index.php/s/XfikGmd2e6pwfaM}} On Parallelizing On-Line Statistics for Stochastic Biological Simulations
Marco Aldinucci
@misc{11:HiBB:simulation,
address = {Bordeaux, France},
author = {Marco Aldinucci},
date-added = {2021-03-18 18:00:00 +0100},
date-modified = {2021-03-18 18:00:00 +0100},
howpublished = {Workshop on High Performance Bioinformatics and Biomedicine (HiBB)},
month = aug,
title = {On Parallelizing On-Line Statistics for Stochastic Biological Simulations},
url = {https://datacloud.di.unito.it/index.php/s/3ijP5zx2ktT6Xab},
year = 2011,
bdsk-url-1 = {https://datacloud.di.unito.it/index.php/s/3ijP5zx2ktT6Xab}} Accelerating code on multi-cores with FastFlow
Marco Aldinucci
@misc{11:EUROPAR:fastflow,
address = {Bordeaux, France},
author = {Marco Aldinucci},
date-added = {2021-03-18 18:00:00 +0100},
date-modified = {2021-03-18 18:00:00 +0100},
howpublished = {Euro-Par 2011},
month = aug,
title = {Accelerating code on multi-cores with FastFlow},
url = {https://datacloud.di.unito.it/index.php/s/Z4RtA9QoxCm829j},
year = 2011,
bdsk-url-1 = {https://datacloud.di.unito.it/index.php/s/Z4RtA9QoxCm829j}} MyMed: Un social network geosensibile per reti fisse e mobili basato sul paradigma Peer-to-Peer
Marco Aldinucci
@misc{11:notteRicercatori:myMed,
address = {Torino, Italy},
annote = {invited},
author = {Marco Aldinucci},
date-added = {2021-03-18 17:00:00 +0100},
date-modified = {2021-03-18 17:00:00 +0100},
howpublished = {La notte dei ricercatori},
month = sep,
note = {invited talk},
title = {MyMed: Un social network geosensibile per reti fisse e mobili basato sul paradigma Peer-to-Peer},
url = {https://datacloud.di.unito.it/index.php/s/QjcSysM93X9DCDn},
year = 2011,
bdsk-url-1 = {https://datacloud.di.unito.it/index.php/s/QjcSysM93X9DCDn}} FastFlow: Performance and Productivity in the Exascale Era
Marco Aldinucci
@misc{11:IBM:fastflow,
address = {Dublin, Ireland},
author = {Marco Aldinucci},
date-added = {2021-03-18 17:00:00 +0100},
date-modified = {2021-03-18 17:00:00 +0100},
howpublished = {IBM Research, Exascale laboratory},
month = oct,
note = {invited talk},
title = {FastFlow: Performance and Productivity in the Exascale Era},
year = 2011,
bdsk-url-1 = {}} FastFlow: Performance and Productivity in the Multicore Era
Marco Aldinucci
@misc{11:FMCO:fastflow,
address = {Torino, Italy},
annote = {invited},
author = {Marco Aldinucci},
date-added = {2021-03-18 17:00:00 +0100},
date-modified = {2021-03-18 17:00:00 +0100},
howpublished = {Formal Methods for Components and Objects (FMCO)},
month = oct,
note = {invited talk},
title = {FastFlow: Performance and Productivity in the Multicore Era},
url = {https://datacloud.di.unito.it/index.php/s/3opjTzm6XRcjYEs},
year = 2011,
bdsk-url-1 = {https://datacloud.di.unito.it/index.php/s/3opjTzm6XRcjYEs}} The camera that makes you beauty (with a novel real-time video-denoiser algorithm)
Marco Aldinucci
@misc{11:TOSM:denoiser,
address = {Torino, Italy},
annote = {invited},
author = {Marco Aldinucci},
date-added = {2021-03-18 17:00:00 +0100},
date-modified = {2021-03-18 17:00:00 +0100},
howpublished = {TOSM expo},
month = nov,
note = {invited talk},
title = {The camera that makes you beauty (with a novel real-time video-denoiser algorithm)},
year = 2011,
bdsk-url-1 = {}} Performance and productivity in the multi-core era: challenges in software engineering and formal methods
Marco Aldinucci
@misc{11:IMT:muticore,
address = {Lucca, Italy},
annote = {invited},
author = {Marco Aldinucci},
date-added = {2021-03-18 17:00:00 +0100},
date-modified = {2021-03-18 17:00:00 +0100},
howpublished = {IMT},
month = nov,
note = {invited talk},
title = {Performance and productivity in the multi-core era: challenges in software engineering and formal methods},
url = {https://datacloud.di.unito.it/index.php/s/Ppok8Cno7enc6QT},
year = 2011,
bdsk-url-1 = {https://datacloud.di.unito.it/index.php/s/Ppok8Cno7enc6QT}} 2010
2010
Efficient Smith-Waterman on multi-core with FastFlow
Marco Aldinucci
@misc{10:PDP:fastflow,
address = {Pisa, Italy},
author = {Marco Aldinucci},
date-added = {2021-03-18 18:00:00 +0100},
date-modified = {2021-03-18 18:00:00 +0100},
howpublished = {PDP 2010: Parallel Distributed and network-based Processing},
month = feb,
title = {Efficient Smith-Waterman on multi-core with FastFlow},
url = {https://datacloud.di.unito.it/index.php/s/2RM8k6j3QGo5R7C},
year = 2010,
bdsk-url-1 = {https://datacloud.di.unito.it/index.php/s/2RM8k6j3QGo5R7C}} FastFlow: a pattern-based programming framework for multicores
Marco Aldinucci
@misc{10:SCHOSS:fastflow,
address = {Wadern, Germania},
annote = {invited},
author = {Marco Aldinucci},
date-added = {2021-03-18 17:00:00 +0100},
date-modified = {2021-03-18 17:00:00 +0100},
howpublished = {seminar 10191},
month = may,
note = {invited talk},
title = {FastFlow: a pattern-based programming framework for multicores},
url = {https://datacloud.di.unito.it/index.php/s/fMpTjXat97kp2Hp},
year = 2010,
bdsk-url-1 = {https://datacloud.di.unito.it/index.php/s/fMpTjXat97kp2Hp}} FastFlow (why we need yet another programming framework)
Marco Aldinucci
@misc{10:belfast:fastflow,
address = {Belfast, United Kingdom},
annote = {invited},
author = {Marco Aldinucci},
howpublished = {Queen's University Belfast},
month = mar,
note = {Guest lecture, 12 March 2010},
title = {{FastFlow} (why we need yet another programming framework)},
year = 2010} 2009
2009
Towards Hierarchical Management of Autonomic Components: a Case Study
Marco Aldinucci
@misc{09:PDP:hierarchical,
address = {Weimar, Germany},
author = {Marco Aldinucci},
date-added = {2021-03-18 21:00:00 +0100},
date-modified = {2021-03-18 21:00:00 +0100},
howpublished = {Euromicro PDP 2009: Parallel Distributed and network-based Processing},
month = feb,
title = {Towards Hierarchical Management of Autonomic Components: a Case Study},
url = {https://datacloud.di.unito.it/index.php/s/KTn3dGwQwbe3bMc},
year = 2009,
bdsk-url-1 = {https://datacloud.di.unito.it/index.php/s/KTn3dGwQwbe3bMc}} Efficient streaming applications on multi-core with FastFlow: the biosequence alignment test-bed
Marco Aldinucci
@misc{09:PARCO:fastflow,
address = {Lyon, France},
author = {Marco Aldinucci},
date-added = {2021-03-18 21:00:00 +0100},
date-modified = {2021-03-18 21:00:00 +0100},
howpublished = {ParCo 2009},
month = sep,
title = {Efficient streaming applications on multi-core with FastFlow: the biosequence alignment test-bed},
url = {https://datacloud.di.unito.it/index.php/s/Aa26ZyrfSB7ZWzj},
year = 2009,
bdsk-url-1 = {https://datacloud.di.unito.it/index.php/s/Aa26ZyrfSB7ZWzj}} STKM on SCA: A unified framework with components, workflows and algorithmic skeletons
Marco Aldinucci
@misc{09:EUROPAR:stkm,
address = {Delft, The Netherlands},
author = {Marco Aldinucci},
date-added = {2021-03-18 21:00:00 +0100},
date-modified = {2021-03-18 21:00:00 +0100},
howpublished = {EuroPar 2009},
month = sep,
title = {STKM on SCA: A unified framework with components, workflows and algorithmic skeletons},
url = {https://datacloud.di.unito.it/index.php/s/KDyBfqmt9HXXwKT},
year = 2009,
bdsk-url-1 = {https://datacloud.di.unito.it/index.php/s/KDyBfqmt9HXXwKT}} 2008
2008
Autonomic components in GCM
Marco Aldinucci
@misc{08:COREGRID:autonomic,
address = {Amsterdam, The Netherland},
annote = {invited},
author = {Marco Aldinucci},
date-added = {2021-03-18 17:00:00 +0100},
date-modified = {2021-03-18 17:00:00 +0100},
howpublished = {CoreGRID Scientific Advisory Board},
month = may,
note = {invited talk},
title = {Autonomic components in GCM},
url = {https://datacloud.di.unito.it/index.php/s/22S3LwsTjbkG3w3},
year = 2008,
bdsk-url-1 = {https://datacloud.di.unito.it/index.php/s/22S3LwsTjbkG3w3}} Towards a Formal Semantics for Autonomic Components
Marco Aldinucci
@misc{08:COREGRID:semantics,
address = {Las Palmas de Gran Canaria, Canary Island, Spain},
author = {Marco Aldinucci},
date-added = {2021-03-18 21:00:00 +0100},
date-modified = {2021-03-18 21:00:00 +0100},
howpublished = {CoreGRID Symposium},
month = aug,
title = {Towards a Formal Semantics for Autonomic Components},
year = 2008,
bdsk-url-1 = {}} The VirtuaLinux Storage Abstraction Layer for Efficient Virtual Clustering
Marco Aldinucci, Massimo Torquati, Marco Vanneschi, Pierfrancesco Zuccato
@misc{08:pdp:virtualinux,
address = {Toulouse, France},
author = {Marco Aldinucci and Massimo Torquati and Marco Vanneschi and Pierfrancesco Zuccato},
howpublished = {16th Euromicro Conference on Parallel, Distributed and Network-based Processing (PDP 2008)},
month = feb,
title = {The {VirtuaLinux} Storage Abstraction Layer for Efficient Virtual Clustering},
year = 2008} 2007
2007
Behavioural skeletons for component autonomic management on grids
Marco Aldinucci
@misc{07:COREGRID:skeletons,
address = {Heraklion, Crete, Greece},
author = {Marco Aldinucci},
date-added = {2021-03-18 21:00:00 +0100},
date-modified = {2021-03-18 21:00:00 +0100},
howpublished = {CoreGRID Workshop on Grid Programming Model, Grid and P2P Systems Architecture, Grid Systems, Tools and Environments},
month = jun,
title = {Behavioural skeletons for component autonomic management on grids},
url = {https://datacloud.di.unito.it/index.php/s/ZpidDEMJo8Pc4nB},
year = 2007,
bdsk-url-1 = {https://datacloud.di.unito.it/index.php/s/ZpidDEMJo8Pc4nB}} GCM Non-Functional Features Advances (Palma MIX)
Marco Aldinucci, Marco Danelutto, Simona Campa, Daniele Laforenza, Nicola Tonellotto, Patrizio Dazzi
@misc{07:gridcomp:palma:gcm,
address = {Palma, Spain},
author = {Marco Aldinucci and Marco Danelutto and Simona Campa and Daniele Laforenza and Nicola Tonellotto and Patrizio Dazzi},
howpublished = {GridCOMP},
note = {Project presentation},
title = {{GCM} Non-Functional Features Advances ({Palma} MIX)},
year = 2007} Adaptive Behavior with GCM
Marco Aldinucci, Marco Danelutto, Simona Campa, Daniele Laforenza, Nicola Tonellotto, Patrizio Dazzi
@misc{07:gridcomp:beijing:gcm,
address = {Beijing, China},
author = {Marco Aldinucci and Marco Danelutto and Simona Campa and Daniele Laforenza and Nicola Tonellotto and Patrizio Dazzi},
howpublished = {GridCOMP},
month = oct,
note = {Project presentation, 31 October 2007},
title = {Adaptive Behavior with {GCM}},
year = 2007} Components, GCM, and Behavioural Skeletons
Marco Aldinucci, Marco Danelutto, Simona Campa, Peter Kilpatrick, Nicola Tonellotto, Patrizio Dazzi
@misc{07:coregrid:belfast:gcm,
address = {Belfast, United Kingdom},
author = {Marco Aldinucci and Marco Danelutto and Simona Campa and Peter Kilpatrick and Nicola Tonellotto and Patrizio Dazzi},
howpublished = {Queen's University Belfast},
month = nov,
note = {Presentation, 22 November 2007},
title = {Components, {GCM}, and Behavioural Skeletons},
year = 2007} VirtuaLinux: Una soluzione open source per il clustering HPC
Marco Aldinucci
@misc{07:NSS:virtualLinux,
address = {Pisa, Italia},
annote = {invited},
author = {Marco Aldinucci},
date-added = {2021-03-18 17:00:00 +0100},
date-modified = {2021-03-18 17:00:00 +0100},
howpublished = {Net & System Security},
month = nov,
note = {invited talk in italian},
title = {VirtuaLinux: Una soluzione open source per il clustering HPC},
url = {https://datacloud.di.unito.it/index.php/s/4SWFntywGgDtoxx},
year = 2007,
bdsk-url-1 = {https://datacloud.di.unito.it/index.php/s/4SWFntywGgDtoxx}} 2006
2006
Autonomic QoS in ASSIST Grid-aware components
Marco Aldinucci
@misc{06:IEEE:qos,
address = {Montb{\'e}liard, France},
author = {Marco Aldinucci},
date-added = {2021-03-18 21:00:00 +0100},
date-modified = {2021-03-18 21:00:00 +0100},
howpublished = {IEEE Euromicro PDP 2006: Parallel Distributed and network-based Processing},
month = feb,
title = {Autonomic QoS in ASSIST Grid-aware components},
url = {https://datacloud.di.unito.it/index.php/s/bcCddKkiF9KsDWM},
year = 2006,
bdsk-url-1 = {https://datacloud.di.unito.it/index.php/s/bcCddKkiF9KsDWM}} Fault-tolerant data sharing for high-level grid programming: a hierarchical storage architecture
Marco Aldinucci
@misc{06:COREGRID:storage,
address = {Krakow, Poland},
author = {Marco Aldinucci},
date-added = {2021-03-18 21:00:00 +0100},
date-modified = {2021-03-18 21:00:00 +0100},
howpublished = {CoreGRID Integration Workshop},
month = oct,
title = {Fault-tolerant data sharing for high-level grid programming: a hierarchical storage architecture},
url = {https://datacloud.di.unito.it/index.php/s/iibrkEtMREHiexA},
year = 2006,
bdsk-url-1 = {https://datacloud.di.unito.it/index.php/s/iibrkEtMREHiexA}} Taming the grid through dynamic adaptation: results and open problems
Marco Aldinucci
@misc{06:LACS:adptation,
address = {San Cristobal de la Laguna, Tenerife, Canarian Islands, Spain},
annote = {invited},
author = {Marco Aldinucci},
date-added = {2021-03-18 17:00:00 +0100},
date-modified = {2021-03-18 17:00:00 +0100},
howpublished = {Last Advances in Computer Science},
month = nov,
note = {invited talk},
title = {Taming the grid through dynamic adaptation: results and open problems},
url = {https://datacloud.di.unito.it/index.php/s/2i8tDjMYEYnZeFo},
year = 2006,
bdsk-url-1 = {https://datacloud.di.unito.it/index.php/s/2i8tDjMYEYnZeFo}} 2005
2005
Building Interoperable Grid-aware ASSIST Applications via Web Services
Marco Aldinucci
@misc{05:PARCO:assist,
address = {Malaga, Spain},
author = {Marco Aldinucci},
date-added = {2021-03-18 21:00:00 +0100},
date-modified = {2021-03-18 21:00:00 +0100},
howpublished = {ParCo 2005},
month = sep,
title = {Building Interoperable Grid-aware ASSIST Applications via Web Services},
url = {https://datacloud.di.unito.it/index.php/s/6rbd7RWE6fbMbsi},
year = 2005,
bdsk-url-1 = {https://datacloud.di.unito.it/index.php/s/6rbd7RWE6fbMbsi}} Towards a distributed scalable data service for the Grid
Marco Aldinucci
@misc{05:PARCO:dataService,
address = {Malaga, Spain},
author = {Marco Aldinucci},
date-added = {2021-03-18 21:00:00 +0100},
date-modified = {2021-03-18 21:00:00 +0100},
howpublished = {ParCo 2005},
month = sep,
title = {Towards a distributed scalable data service for the Grid},
url = {https://datacloud.di.unito.it/index.php/s/zQMGHBAWRLdKsbH},
year = 2005,
bdsk-url-1 = {https://datacloud.di.unito.it/index.php/s/zQMGHBAWRLdKsbH}} Dynamic reconfiguration of grid-aware applications in ASSIST
Marco Aldinucci
@misc{05:EUROPAR:assist,
address = {Lisbon, Portugal},
author = {Marco Aldinucci},
date-added = {2021-03-18 21:00:00 +0100},
date-modified = {2021-03-18 21:00:00 +0100},
howpublished = {Euro-Par 2005},
month = sep,
title = {Dynamic reconfiguration of grid-aware applications in ASSIST},
url = {https://datacloud.di.unito.it/index.php/s/nPESNMGWirAYEFB},
year = 2005,
bdsk-url-1 = {https://datacloud.di.unito.it/index.php/s/nPESNMGWirAYEFB}} 2004
2004
Optimization techniques for implementing parallel skeletons in Grid
Marco Aldinucci
@misc{04:CMPP:skeletons,
address = {Stirling, Scotland},
author = {Marco Aldinucci},
date-added = {2021-03-18 21:00:00 +0100},
date-modified = {2021-03-18 21:00:00 +0100},
howpublished = {CMPP 2004 (in conjunction with MPC 04)},
month = jul,
title = {Optimization techniques for implementing parallel skeletons in Grid},
url = {https://datacloud.di.unito.it/index.php/s/MozKWbWPCxko2CX},
year = 2004,
bdsk-url-1 = {https://datacloud.di.unito.it/index.php/s/MozKWbWPCxko2CX}} Accelerating Apache farms through ad-HOC distributed scalable object repository
Marco Aldinucci
@misc{04:EUROPAR:apache,
address = {Pisa, Italy},
author = {Marco Aldinucci},
date-added = {2021-03-18 21:00:00 +0100},
date-modified = {2021-03-18 21:00:00 +0100},
howpublished = {Euro-Par 2004},
month = sep,
title = {Accelerating Apache farms through ad-HOC distributed scalable object repository},
url = {https://datacloud.di.unito.it/index.php/s/zLXTRRQb2rJNyDy},
year = 2004,
bdsk-url-1 = {https://datacloud.di.unito.it/index.php/s/zLXTRRQb2rJNyDy}} Rendering Grid Heterogeneity Harmless
Marco Aldinucci
@misc{04:FGG:heterogeneity,
address = {Dagstuhl, Germany},
annote = {invited},
author = {Marco Aldinucci},
date-added = {2021-03-18 17:00:00 +0100},
date-modified = {2021-03-18 17:00:00 +0100},
howpublished = {Seminar 04451 -- Future Generation Grids},
month = nov,
note = {invited talk},
title = {Rendering Grid Heterogeneity Harmless},
url = {https://datacloud.di.unito.it/index.php/s/ofZiZrd8YpWgqZD},
year = 2004,
bdsk-url-1 = {https://datacloud.di.unito.it/index.php/s/ofZiZrd8YpWgqZD}} 2003
2003
eskimo: experimenting skeletons on the shared address model
Marco Aldinucci
@misc{03:HLPP:eskimo,
address = {Paris, France},
author = {Marco Aldinucci},
date-added = {2021-03-18 22:00:00 +0100},
date-modified = {2021-03-18 22:00:00 +0100},
howpublished = {HLPP 2003},
month = jun,
title = {eskimo: experimenting skeletons on the shared address model},
url = {https://datacloud.di.unito.it/index.php/s/noo2N7HBKWm2nMH},
year = 2003,
bdsk-url-1 = {https://datacloud.di.unito.it/index.php/s/noo2N7HBKWm2nMH}} ASSIST demo: a high level, high performance, portable, structured parallel programming environment at work
Marco Aldinucci
@misc{03:EUROPAR:assist,
address = {Austria},
author = {Marco Aldinucci},
date-added = {2021-03-18 21:00:00 +0100},
date-modified = {2021-03-18 21:00:00 +0100},
howpublished = {9th Intl Euro-Par 2003: Parallel and Distributed Computing},
month = aug,
title = {ASSIST demo: a high level, high performance, portable, structured parallel programming environment at work},
url = {https://datacloud.di.unito.it/index.php/s/fmEcpJDGD4iitia},
year = 2003,
bdsk-url-1 = {https://datacloud.di.unito.it/index.php/s/fmEcpJDGD4iitia}} An operational semantics for skeletons
Marco Aldinucci
@misc{03:PARCO:skeletons,
address = {Dresden, Germany},
author = {Marco Aldinucci},
date-added = {2021-03-18 21:00:00 +0100},
date-modified = {2021-03-18 21:00:00 +0100},
howpublished = {ParCo 2003},
month = sep,
title = {An operational semantics for skeletons},
url = {https://datacloud.di.unito.it/index.php/s/xDnFpT6RM9LHAEd},
year = 2003,
bdsk-url-1 = {https://datacloud.di.unito.it/index.php/s/xDnFpT6RM9LHAEd}} A framework for experimenting with structured parallel programming environment design
Marco Aldinucci
@misc{03:PARCO:framework,
address = {Dresden, Germany},
author = {Marco Aldinucci},
date-added = {2021-03-18 21:00:00 +0100},
date-modified = {2021-03-18 21:00:00 +0100},
howpublished = {ParCo 2003},
month = sep,
title = {A framework for experimenting with structured parallel programming environment design},
url = {https://datacloud.di.unito.it/index.php/s/FkGk3giN8MiGMwo},
year = 2003,
bdsk-url-1 = {https://datacloud.di.unito.it/index.php/s/FkGk3giN8MiGMwo}} 2000
2000
The META transformation tool for skeleton-based language
Marco Aldinucci
@misc{00:CMPP:META,
address = {Ponte de Lima, Portugal},
author = {Marco Aldinucci},
date-added = {2021-03-18 22:00:00 +0100},
date-modified = {2021-03-18 22:00:00 +0100},
howpublished = {CMPP 2000},
month = jul,
title = {The META transformation tool for skeleton-based language},
url = {https://datacloud.di.unito.it/index.php/s/e9L5fWDo9q5fmw4},
year = 2000,
bdsk-url-1 = {https://datacloud.di.unito.it/index.php/s/e9L5fWDo9q5fmw4}}