Robert Birke

@article{Birke2026TMPDiff,
  author        = {Lewandowski, Basile and Kurz, Simon and Shankar, Aditya and Birke, Robert and Chen, Jian-Jia and Chen, Lydia Y.},
  title         = {{TMPDiff}: Temporal Mixed-Precision for Diffusion Models},
  journal       = {arXiv preprint arXiv:2603.14062},
  year          = {2026},
  eprint        = {2603.14062},
  archiveprefix = {arXiv},
  primaryclass  = {cs.CV},
  doi           = {10.48550/arXiv.2603.14062},
  url           = {https://arxiv.org/abs/2603.14062}
}

@inproceedings{Vaccaro2026DigitalTwins,
	abstract = {Digital Twins are increasingly adopted to support the operational management of complex
and energy-intensive infrastructures, yet their practical implementation often remains
fragmented across heterogeneous platforms and proprietary monitoring systems. In data
center environments, this fragmentation limits the interpretability of real-time data and
weakens the connection between operational measurements, spatial context, and asset-
level decision-making. This paper presents a BIM-based Digital Twin architecture developed
for the HPC4AI (High-Performance Computing for Artificial Intelligence) data center, hosted
at the University of Turin's Computer Science Department, within the framework of the
DYMAN project, which aims to develop advanced solutions to reduce the impact of thermal
energy management in high-performance computing installations and data server rooms.
The proposed approach integrates real-time IoT monitoring with Building Information
Modelling through a modular layered architecture that decouples sensing, time-series data
management, and digital representation. Sensor data are stored in a time-series database
and synchronized into the BIM environment via automated Dynamo workflows and
embedded Python scripts, where shared parameters enable semantic mapping to building
elements. This supports near real-time spatial contextualization, rule-based alerting, and
native BIM visualisation without reliance on external dashboards. The case study
demonstrates how the BIM model evolves from a static geometric artifact into an
operational interface capable of supporting situational awareness, diagnostics, and
integration with predictive analytics and energy optimization strategies. The paper discusses
system architecture, data synchronization mechanisms, and transferability to other
mission-critical infrastructures, highlighting the role of data-driven integration strategies in
advancing scalable and interoperable Digital Twins for smart and sustainable built
environments.},
	address = {Cambridge, UK},
	author = {Vaccaro, Viviana and Tagliabue, Lavinia Chiara and Birke, Robert and Aldinucci, Marco and Meschini, Silvia},
	booktitle = {Proc. of Smart and Sustainable Built Environment Conference Series},
	date-modified = {2026-09-01 23:50:08 +0200},
	issn = {2979-1804},
	keywords = {dyman, hpc4ai, DCs},
	month = {jul},
	pages = {1-12},
	title = {Digital Twins for Mission-Critical Infrastructures: A {BIM}-Based Framework for Real-Time Monitoring and Decision Support},
	url ={https://dyman-project.eu/download/digital-twins-for-mission-critical-infrastructures-a-bim-based-framework-for-real-time-monitoring-and-decision-support/?wpdmdl=10085&refresh=6a974a29bb57b1788299817},
	year = {2026}}

@inproceedings{Tagliabue2026EnhancingDatacenter,
	abstract = {Digital technologies are transforming how we manage and repurpose the built environment, enhancing energy performance, system reliability, and user wellbeing. Among these innovations, datacenters stand out as essential infrastructures due to their scale, criticality, and intensive energy usage. Traditionally housed in purpose-built facilities, datacenters are increasingly being considered for integration into existing
	buildings, including those of historical or industrial value. This adaptive reuse offers an opportunity to reconcile heritage conservation with contemporary needs for connectivity, sustainability, and digital resilience. This paper presents two case studies exploring the integration of digital infrastructure into a historic building and a newly constructed facility. Both projects address key challenges related to energy efficiency, spatial constraints, and architectural adaptation. Emphasis is placed on strategies for heat recovery, enabling the reuse of waste heat for other compatible functions, and on the deployment of Digital Twins to simulate, monitor, and optimize performance in real time. These examples demonstrate how datacenters can catalyze sustainable regeneration while maintaining the cultural value of heritage sites. By aligning digital innovation with architectural reuse, the projects highlight a replicable model for embedding AI-driven systems into the future of built heritage.},
	address = {Rome, Italy},
	author = {Tagliabue, Lavinia Chiara and Aldinucci, Marco and Vaccaro, Viviana and Meschini, Silvia and Birke, Robert},
	booktitle = {Proceedings of the Stati Generali del Patrimonio Industriale 2026},
	date-modified = {2026-08-14 12:33:52 +0200},
	editor = {Curr{\`a}, Edoardo and Fatiguso, Fabio and Guida, Antonella and Monte, Antonio and Docci, Marina and Bernardo, Graziella and Cantatore, Elena and Mattioli, Alessandro and Menichelli, Claudio},
	isbn = {978-88-492-5487-7},
	keywords = {DCs, dyman},
	publisher = {Gangemi Editore},
	title = {Enhancing Datacenter Integration in the Built Environment through Digital Twins},
	url = {https://iris.unito.it/handle/2318/2124617},
	year = {2026},
	bdsk-url-1 = {https://iris.unito.it/handle/2318/2124617}}

@inproceedings{11626216,
	abstract = {The rapid expansion of data centers driven by cloud computing, high performance computing, and artificial intelligence has intensified the need for reliable frameworks to assess and communicate their environmental sustainability. Among existing tools, the Leadership in Energy and Environmental Design (LEED) certification has emerged as the most widely adopted green building rating system applied to data centers. However, LEED was originally developed for conventional building typologies, raising questions regarding its adequacy for capturing the specific environmental characteristics of data center facilities. This paper critically examines the application of LEED certification to data centers, focusing on LEED O+M: Data Centers (v4). The study combines a review of official USGBC guidance with a comparative analysis of multiple LEED-certified data center scorecards awarded between 2015 and 2024. The distribution of points across certification categories is analyzed to identify recurring patterns, dominant credit areas, and structural limitations. The results show that LEED certification for data centers is strongly driven by operational energy performance, with the Energy & Atmosphere category consistently accounting for the largest share of awarded points in Gold and Platinum projects. Credits related to commissioning, energy monitoring, and performance optimization play a decisive role in achieving high certification levels. Conversely, categories addressing materials, resource use, and long-term environmental impacts remain marginal, while occupant-centric criteria retain significant weight despite limited relevance to low-occupancy data center environments. The findings indicate that LEED is an effective framework for operational sustainability benchmarking and governance in data centers but provides a partial representation of their overall environmental performance. The paper concludes by discussing the need for integrating certification frameworks with complementary analytical approaches capable of addressing life-cycle impacts and technological dynamics, thereby enhancing the robustness of sustainability assessment for data centers.},
	author = {Tagliabue, Lavinia Chiara and Vaccaro, Viviana and Zichi, Alessandro and Birke, Robert and Aldinucci, Marco and Meschini, Silvia},
	booktitle = {2026 34th Euromicro International Conference on Parallel, Distributed, and Network-Based Processing (PDP)},
	date-modified = {2026-08-14 12:10:40 +0200},
	doi = {10.23919/PDPCPS00030.2026.00047},
	keywords = {DCs, dyman},
	month = {March},
	pages = {300-307},
	publisher = {{IEEE}},
	title = {LEED Certification and Data Centers: A Critical Assessment of Building Sustainability Rating Systems for HPC Facilities},
	year = {2026},
	bdsk-url-1 = {https://doi.org/10.23919/PDPCPS00030.2026.00047}}

@inproceedings{11626207,
	abstract = {High-performance computing (HPC) and artificial intelligence (AI) data centers are increasingly recognized as critical infrastructures for scientific research, yet they are also associated with significant environmental burdens due to their high energy intensity and rapid technological obsolescence. Traditional performance metrics such as Power Usage Effectiveness (PUE) primarily capture operational efficiency and fail to account for embodied impacts related to construction, equipment manufacturing, and component replacement. This study presents a preliminary Life Cycle Assessment (LCA) of the HPC4AI data center at the University of Turin, conducted in accordance with ISO 14040--44 [1] and structured using the EN 15978 modular framework. The assessment adopts a screening-level, attributional LCA with a 30-year reference study period, integrating primary inventory data derived from seven detailed design and equipment schedules covering architectural elements, IT infrastructure, electrical systems, lighting, and cooling equipment. Life-cycle impacts are quantified for the modules A1--A3 (product stage), B4 (replacement due to technological obsolescence), and B6 (operational energy use), while other modules are structurally included as placeholders. Global warming potential (GWP) is used as the impact indicator, with multiple electricity mix scenarios applied to evaluate operational sensitivity. Results indicate that operational electricity use (Module B6) is the dominant contributor to life-cycle GWP, confirming findings from recent HPC and AI data center LCAs. However, embodied impacts associated with IT and cooling equipment manufacturing and replacement (A1--A3 + B4) represent a substantial secondary contribution, with replacement impacts of comparable magnitude to initial construction. A component-level ranking reveals that servers, UPS systems, and core cooling equipment are the principal sources of embodied emissions, while architectural components contribute marginally. The study highlights the critical role of electricity decarbonization, hardware lifetime management, and circular strategies in reducing the environmental footprint of academic HPC facilities. Despite its screening nature, the LCA provides a robust baseline for future refined assessments and supports the integration of life-cycle thinking into the design and governance of research-oriented data centers.},
	author = {Tagliabue, Lavinia Chiara and Vaccaro, Viviana and Zichi, Alessadro and Birke, Robert and Aldinucci, Marco},
	booktitle = {2026 34th Euromicro International Conference on Parallel, Distributed, and Network-Based Processing (PDP)},
	date-modified = {2026-08-14 12:10:48 +0200},
	doi = {10.23919/PDPCPS00030.2026.00045},
	keywords = {DCs, dyman},
	month = {March},
	pages = {285-292},
	publisher = {{IEEE}},
	title = {Preliminary Life Cycle Assessment of a HighPerformance Computing Data Center: A Case Study of the HPC4AI Facility at the University of Turin},
	year = {2026},
	bdsk-url-1 = {https://doi.org/10.23919/PDPCPS00030.2026.00045}}

@inproceedings{11626213,
	abstract = {Data centers (DCs) represent one of the most energy-intensive and operationally critical categories of buildings, supporting essential digital services such as cloud computing, large-scale data storage, and artificial intelligence applications. Their operation is characterized by complex interdependencies between information technology loads, cooling infrastructures, power distribution systems, and environmental conditions. Ensuring efficiency, reliability, and resilience under these constraints remains a significant challenge for operators and designers alike. In recent years, the concept of Digital Twin (DT) has emerged as a promising paradigm to address this complexity by enabling a continuously updated digital representation of physical systems, enriched with real-time data and analytical capabilities. Despite increasing attention in both academia and industry, DT solutions for buildings and DCs remain largely fragmented. Existing approaches often focus on individual aspects such as visualization, simulation, or monitoring dashboards, while failing to provide robust integration between design data, operational measurements, and control systems. A recurring limitation concerns the lack of interoperability between Building Information Modeling (BIM) data and Internet of Things (IoT) infrastructures, which prevents the construction of coherent and scalable DT systems. This paper proposes a reference architecture for DT applied to DCs, grounded in the integration of BIM-based representations, real-time IoT data acquisition, analytical models, and control mechanisms. The architecture adopts a layered perspective, separating concerns related to data management, semantic integration, modeling, analytics, and actuation, while maintaining consistency across these layers. Informed by recent reviews on building-scale DT, the proposed approach addresses the well-known bottlenecks associated with BIM usage and data acquisition. The contribution of this work lies in providing an architectural blueprint able to support interoperability and reuse, offering a structured foundation for the development of advanced DT solutions in DC environments, and complemented by practical metrics that make DT implementations assessable beyond project-specific prototypes.},
	author = {Meschini, Silvia and Gasbarri, Paola and Tagliabue, Lavinia Chiara and Vaccaro, Viviana and Birke, Robert and Aldinucci, Marco},
	booktitle = {34th Euromicro International Conference on Parallel, Distributed, and Network-Based Processing {(PDP)}},
	date-modified = {2026-08-14 12:10:56 +0200},
	doi = {10.23919/PDPCPS00030.2026.00048},
	keywords = {DCs, dyman},
	pages = {308-315},
	publisher = {{IEEE}},
	title = {Digital Twin for DCs: a Reference Architecture Based On BIM, IoT and Analytics Control},
	year = {2026},
	bdsk-url-1 = {https://doi.org/10.23919/PDPCPS00030.2026.00048}}

@inproceedings{2026:icsuperhype,
	abstract = {Hypergraphs are graph generalizations with key applications in domains such as healthcare, where strict data privacy requirements apply, or bioinformatics, where testing new compounds is costly. However, due to their combinatorial nature, hypergraph representations are often either intractable or lead to significant information loss. For this reason, research into hypergraph synthesis is limited, and state-of-the-art approaches yield poor generation quality in terms of overall structural patterns and graph-level validity. To address such shortcomings, we introduce SuperHype, an exact and tractable hypergraph diffusion model. The core of SuperHype is the graph-superposition decomposition, a novel representation that embeds a hypergraph into a multi-layer graph, enabling a tractable representation with no loss of generalization. To generate new samples from such representations, we introduce a Graph-Superposition Transformer that treats the superposition as an interconnected sequence of layers. Moreover, we enhance the model's performance by incorporating hypergraph-specific auxiliary features and aggregating indirect node interactions via triplet pooling. Our evaluation across five datasets shows that SuperHype generally reproduces local and global connectivity patterns with superior fidelity compared to state-of-the-art baselines.},
	author = {Lucas Gantes and Abele M{\u{a}}lan and Roberto Gheda and Robert Birke and Lydia Chen},
	booktitle = {Forty-third International Conference on Machine Learning},
	keywords = {dyman, ai},
	title = {SuperHype: Hypergraph Generation via Graph-Superposition Decomposition},
	url = {https://openreview.net/forum?id=VasXWJMqSp},
	year = {2026},
	bdsk-url-1 = {https://openreview.net/forum?id=VasXWJMqSp}}

@article{2026:fcgs:performance,
	author = {Garcia, Adriano Marques and Malenza, Giulio and Birke, Robert and Aldinucci, Marco},
	doi = {10.1016/J.FUTURE.2025.108242},
	journal = {Future Gener. Comput. Syst.},
	keywords = {icsc, dare},
	pages = {108242},
	title = {Inference performance of large language models on a 64-core {RISC-V} {CPU} with silicon-enabled vectors},
	volume = {177},
	year = {2026},
	bdsk-url-1 = {https://doi.org/10.1016/J.FUTURE.2025.108242}}

@article{25:FGCS:TEEs-Performance,
	abstract = {Data confidentiality is crucial when processing sensitive information, often limiting user interactions and shared computing services like the cloud. While Trusted Execution Environments (TEEs) offer a means to ensure privacy in untrusted environments, they frequently introduce significant computational overhead. DNA alignment, a key step in bioinformatics workflows, is privacy-sensitive and computationally intensive. Given its parallelizable nature, it is a compelling case study for evaluating the performance impact and scalability of various TEEs. This study assesses three TEEs -- Intel SGX, Intel TDX, and AMD SEV-SNP -- by evaluating their overhead through real-world bioinformatics workloads and system-level microbenchmarks. Our evaluation shows that SGX-based solutions incur substantial overhead, particularly for small workloads, with slowdowns ranging from 283% to 1971% compared to native execution. The overhead is reduced for larger workloads, ranging from 15% to 57%. In contrast, TDX and SEV-SNP offer significantly improved performance: TDX limits overhead to 73% for small and to 9% for large workloads, while SEV-SNP incurs at most 67% and 29%, respectively. Importantly, SEV-SNP demonstrates better scalability than TDX, a result supported by microbenchmark analysis showing more efficient thread creation and scheduling. Conversely, TDX shows more efficient memory utilization, underscoring distinct overhead sources among the evaluated TEE architectures.},
	author = {Lorenzo Brescia and Iacopo Colonnelli and Robert Birke and Valerio Schiavoni and Pascal Felber and Marco Aldinucci},
	doi = {10.1016/j.future.2025.108031},
	issn = {0167-739X},
	journal = {Future Generation Computer Systems},
	keywords = {icsc, confidential},
	pages = {108031},
	title = {A comprehensive performance evaluation of TEEs for confidential DNA alignment},
	url = {https://www.sciencedirect.com/science/article/pii/S0167739X25003267},
	year = {2026},
	bdsk-url-1 = {https://www.sciencedirect.com/science/article/pii/S0167739X25003267},
	bdsk-url-2 = {https://doi.org/10.1016/j.future.2025.108031}}

@inproceedings{Santimaria2026-TMT-hipes-25,
  title = {Thread Monitoring Tool: Transparent Characterization of Threading Patterns with eBPF},
  isbn = {9783032191441},
  issn = {1611-3349},
  url = {http://dx.doi.org/10.1007/978-3-032-19144-1_20},
  doi = {10.1007/978-3-032-19144-1_20},
  booktitle = {Lecture Notes in Computer Science},
  publisher = {Springer Nature Switzerland},
  author = {Santimaria,  Marco Edoardo and Marques Garcia,  Adriano and Malenza,  Giulio and Monaldi,  Stefano and Aldinucci,  Marco and Birke,  Robert},
  year = {2026},
  keywords = {icsc,dare},
  pages = {253–264},
  journal = {Lecture Notes in Computer Science},
  abstract = {Controlling the number of threads in parallel computation is a key tuning factor for performance. If too few, the program does not leverage all the available hardware resources. If too many, thread management overhead and bottleneck-resource congestion slow down execution. A good rule-of-thumb starting value is often one software thread for each hardware core. However, complex software may use multiple parallelism at different (nested) levels, making it difficult to predict what is created during execution. Moreover, classic monitoring tools only sample the process tree at given time instants, which misses any thread not yet created or already destroyed before sampling happens. Here, we present a thread monitoring tool that traces kernel events for continuous tracking of threads spawned and joined by a program during its execution with little to no overhead. Evaluation of simple applications running with different parallel backends, as well as PyTorch with the LAMA-3.2-1B and Bzip2 running with two different parallel backends, shows the tool’s capabilities.},
}

@inproceedings{bridda2026t2f,
  author    = {Consuelo Bridda and Doriana Medi{\'c} and Lavinia Chiara Tagliabue and Robert Birke and Marco Aldinucci},
  title     = {T2F: Forecasting Irregularly Sampled Time Series via Time Features},
  booktitle = {Accepted for publication in: 2026 IEEE International Symposium on Precision Clock Synchronization for Measurement, Control, and Communication (ISPCS)},
  year      = {2026},
  address   = {Brescia, Italy},
  keywords  = {ai}
}

@article{Malenza2025OpenFOAMCpp,
  author        = {Malenza, Giulio and Stabile, Giovanni and Spiga, Filippo and Birke, Robert and Aldinucci, Marco},
  title         = {Building an Accelerated {OpenFOAM} Proof-of-Concept Application using Modern {C++}},
  journal       = {arXiv preprint arXiv:2507.18268},
  year          = {2025},
  eprint        = {2507.18268},
  archiveprefix = {arXiv},
  primaryclass  = {cs.DC},
  doi           = {10.48550/arXiv.2507.18268},
  url           = {https://arxiv.org/abs/2507.18268}
}

@inproceedings{gheda2025collaborative,
	abstract = {Bayesian Network models are a powerful tool to collaboratively optimize production processes in various manufacturing industries. When interacting, collaborating parties must preserve their business secrets by maintaining the confidentiality of their model structures and parameters. While most realistic industry scenarios involve hybrid settings, handling both discrete and continuous data, current state-ofthe-art methods for collaborative and confidential inference only support discrete data and have high communication costs. In a centralized setting, Junction Trees enable efficient inference even in hybrid scenarios without discretizing continuous variables, but no extension for collaborative and confidential scenarios exists. To address this research gap, we introduce Hybrid CCJT, the first framework for confidential multiparty inference in hybrid domains with semi-honest, non-colluding adversaries, comprising: (i) a method to construct a strongly-rooted Junction Tree across collaborating parties through a novel construct of interface cliques; and, (ii) a protocol for confidential inference built upon multiparty computation primitives comprising a one-time alignment phase and a belief propagation system for combining the inference results across the Junction Tree cliques. Extensive evaluation on nine datasets shows that Hybrid CCJT improves the predictive accuracy of continuous target variables by 32% on average compared to the state-of-the-art, while reducing communication costs by a median 10.4× under purely discrete scenarios.},
	author = {Gheda, Roberto and M{\u{a}}lan, Abele and Guzella, Thiago and Lancia, Carlo and Birke, Robert and Chen, Lydia Y.},
	booktitle = {Proceedings of The Thirty-ninth Annual Conference on Neural Information Processing Systems},
	keywords = {dyman, icsc, ai},
	location = {San Diego, CA, USA},
	title = {Collaborative and Confidential Junction Trees for Hybrid Bayesian Networks},
	url = {https://openreview.net/forum?id=ehfIVCHxkX},
	year = {2025},
	bdsk-url-1 = {https://iris.unito.it/retrieve/04313a93-09f6-4bb3-98e8-d2e8225cf7f3/22683_Collaborative_and_Confid.pdf}}

@article{25:TPDS:PPGAIA,
	author = {Malenza, Giulio and Cesare, Valentina and Santimaria, Marco Edoardo and Birke, Robert and Vecchiato, Alberto and Becciani, Ugo and Aldinucci, Marco},
	doi = {10.1109/tpds.2025.3591452},
	issn = {2161-9883},
	journal = {IEEE Transactions on Parallel and Distributed Systems},
	keywords = {astro, HPC, dyman, icsc},
	pages = {1--12},
	publisher = {Institute of Electrical and Electronics Engineers (IEEE)},
	title = {Performance Portability Assessment in Gaia},
	url = {http://dx.doi.org/10.1109/TPDS.2025.3591452},
	year = {2025},
	bdsk-url-1 = {http://dx.doi.org/10.1109/TPDS.2025.3591452}}

@article{25:RISCV64:HLPP,
	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 AI applications in parallel on new architectures to fully harness their maximum performance remains a challenge. In this study, we investigate the impact of model parallelism on 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 data locality stemming from splitting and assigning parts of the model to specific (groups of) cores handling dependencies via 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 approach based on data parallelism 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. In our experiments, using model parallelism improved up to 8.4 times the performance over the native PyTorch parallelism.},
	author = {Malenza, Giulio and Garcia, Adriano Marques and Birke, Robert and Benini, Luca and Aldinucci, Marco},
	date = {2025/06/30},
	date-added = {2025-07-01 00:02:01 +0200},
	date-modified = {2025-07-01 00:02:13 +0200},
	doi = {10.1007/s10766-025-00802-6},
	isbn = {1573-7640},
	journal = {International Journal of Parallel Programming},
	keywords = {icsc, eupilot, dyman, riscv},
	number = {4},
	pages = {27},
	title = {Analysis of Model Parallelism for AI Applications on a 64-core RV64 Server CPU},
	volume = {53},
	year = {2025},
	bdsk-url-1 = {https://doi.org/10.1007/s10766-025-00802-6}}

@article{10.1145/3729215,
	abstract = {Recent trends in deep learning (DL) have made hardware accelerators essential for various high-performance computing (HPC) applications, including image classification, computer vision, and speech recognition. This survey summarizes and classifies the most recent developments in DL accelerators, focusing on their role in meeting the performance demands of HPC applications. We explore cutting-edge approaches to DL acceleration, covering not only GPU- and TPU-based platforms but also specialized hardware such as FPGA- and ASIC-based accelerators, Neural Processing Units, open hardware RISC-V-based accelerators, and co-processors. This survey also describes accelerators leveraging emerging memory technologies and computing paradigms, including 3D-stacked Processor-In-Memory, non-volatile memories like Resistive RAM and Phase Change Memories used for in-memory computing, as well as Neuromorphic Processing Units, and Multi-Chip Module-based accelerators. Furthermore, we provide insights into emerging quantum-based accelerators and photonics. Finally, this survey categorizes the most influential architectures and technologies from recent years, offering readers a comprehensive perspective on the rapidly evolving field of deep learning acceleration.},
	address = {New York, NY, USA},
	articleno = {286},
	author = {Silvano, Cristina and Ielmini, Daniele and Ferrandi, Fabrizio and Fiorin, Leandro and Curzel, Serena and Benini, Luca and Conti, Francesco and Garofalo, Angelo and Zambelli, Cristian and Calore, Enrico and Schifano, Sebastiano and Palesi, Maurizio and Ascia, Giuseppe and Patti, Davide and Petra, Nicola and De Caro, Davide and Lavagno, Luciano and Urso, Teodoro and Cardellini, Valeria and Cardarilli, Gian Carlo and Birke, Robert and Perri, Stefania},
	doi = {10.1145/3729215},
	issn = {0360-0300},
	issue_date = {November 2025},
	journal = {ACM Comput. Surv.},
	keywords = {icsc, riscv},
	month = jun,
	number = {11},
	numpages = {39},
	publisher = {Association for Computing Machinery},
	title = {A Survey on Deep Learning Hardware Accelerators for Heterogeneous HPC Platforms},
	url = {https://doi.org/10.1145/3729215},
	volume = {57},
	year = {2025},
	bdsk-url-1 = {https://doi.org/10.1145/3729215}}

@inproceedings{lek:SAM:ICMLW25,
	author = {Lek, Gert and Zhu, Chaoyi and Chen, Pin-Yu and Birke, Robert and Chen, Lydia Y.},
	booktitle = {ICML 2025 Workshop DIG-BUG},
	keywords = {icsc},
	title = {Detective {SAM}: Adapting {SAM} to Localize Diffusion-based Forgeries via Embedding Artifacts},
	url = {https://openreview.net/forum?id=GPpFTmGLgT},
	year = {2025},
	bdsk-url-1 = {https://openreview.net/forum?id=GPpFTmGLgT}}

@inproceedings{Zhao:Tabula,
	author = {Zhao, Zilong and Birke, Robert and Chen, Lydia Y.},
	booktitle = {Advances in Knowledge Discovery and Data Mining - 29th Pacific-Asia Conference on Knowledge Discovery and Data Mining, {PAKDD} 2025},
	doi = {10.1007/978-981-96-8186-0\_20},
	pages = {247--259},
	publisher = {Springer},
	series = {Lecture Notes in Computer Science},
	title = {TabuLa: Harnessing Language Models for Tabular Data Synthesis},
	volume = {15874},
	year = {2025},
	bdsk-url-1 = {https://doi.org/10.1007/978-981-96-8186-0%5C_20}}

@inproceedings{vaccar0:pdp25,
	abstract = {The HPC4AI datacenter hosted at the Computer Science Department of the University of Turin was born to address the exponentially increasing computing needs of cross-disciplinary research on AI. To address the needs of modern AI, HPC4AI rethinks the traditional usage of Cloud and HPC systems where the Cloud provides a modern interface for HPC, and HPC serves as an accelerator for the cloud. To date, it has supported 40+ research projects across a broad range of domains, from astronomy and medicine to human sciences. Furthermore, it acts as an R\&D platform to study, develop, and test new datacenter technologies. As such, it hosts a zoo of exotic computing platforms and the first prototype of two-phase evaporative server cooling. This work describes operating and managing HPC4AI with its challenges and lessons learned, with an analysis of key opportunities for digital twins.},
	author = {Vaccaro, Viviana and Birke, Robert and Tagliabue, Lavinia Chiara and Rabellino, Sergio and Aldinucci, Marco},
	booktitle = {2025 33rd Euromicro International Conference on Parallel, Distributed, and Network-Based Processing (PDP)},
	date-modified = {2026-08-14 12:13:05 +0200},
	doi = {10.1109/PDP66500.2025.00077},
	keywords = {DCs, icsc, dyman},
	pages = {499--505},
	title = {HPC4AI@UNITO: A Use Case For Datacenter Digital Twin},
	url = {https://iris.unito.it/retrieve/652d1a4f-69fd-4690-a3b0-9d5723b4db4b/PDP25_DT4DC_HPC4AI.pdf},
	year = {2025},
	bdsk-url-1 = {https://ieeexplore.ieee.org/document/10974756},
	bdsk-url-2 = {https://doi.org/10.1109/PDP66500.2025.00077}}

@inproceedings{zhu2025:iclr,
	abstract = {Synthetic data offers alternatives for data augmentation and sharing. Till date, it remains unknown how to use watermarking techniques to trace and audit synthetic tables generated by tabular diffusion models to mitigate potential misuses. In this paper, we design TabWak, the first watermarking method to embed invisible signatures that control the sampling of Gaussian latent codes used to synthesize table rows via the diffusion backbone. TabWak has two key features. Different from existing image watermarking techniques, TabWak uses self-cloning and shuffling to embed the secret key in positional information of random seeds that control the Gaussian latents, allowing to use different seeds at each row for high inter-row diversity and enabling row-wise detectability. To further boost the robustness of watermark detection against post-editing attacks, TabWak uses a valid-bit mechanism that focuses on the tail of the latent code distribution for superior noise resilience. We provide theoretical guarantees on the row diversity and effectiveness of detectability. We evaluate TabWak on five datasets against baselines to show that the quality of watermarked tables remains nearly indistinguishable from non-watermarked tables while achieving high detectability in the presence of strong post-editing attacks, with a 100\% true positive rate at a 0.1\% false positive rate on synthetic tables with fewer than 300 rows. Our code is available at the following repository https://github.com/chaoyitud/TabWak.},
	author = {Zhu, Chaoyi and Tang, Jiayi and Galjaard, Jeroen and Chen, Pin-Yu and Birke, Robert and Bos, Cornelis and Chen, Lydia Y.},
	booktitle = {The Thirteenth International Conference on Learning Representations},
	keywords = {icsc, dyman},
	title = {TabWak: A Watermark for Tabular Diffusion Models},
	url = {https://openreview.net/forum?id=71pur4y8gs},
	year = {2025},
	bdsk-url-1 = {https://openreview.net/forum?id=71pur4y8gs}}

@inproceedings{23:itadata:mittone:fonio,
	abstract = {Federated learning (FL) is a distributed machine learning paradigm allowing cooperative model training between multiple parties while maintaining local data privacy. FL can be deployed at various scales, ranging from thousands of low-end devices (e.g., smartphones) to just a few high-performance infrastructures (e.g., HPCs), raising critical concerns about the scalability of state-of-the-art FL frameworks. This preliminary study evaluates the scaling performance of a representative FL framework, i.e., Flower, in high-performance controlled environments, providing insights on such frameworks from a computational performance point of view. Two public Top500 pre- exascale HPC infrastructures are exploited to obtain reliable and comparable results: Leonardo and MareNostrum5. Our findings suggest that the design of current FL frameworks, and especially their communication backends, may overlook computational performance, leading to poor scaling in large-scale scenarios.},
	address = {Turin, Italy},
	author = {Mittone, Gianluca and Fonio, Samuele and Birke, Robert and Aldinucci, Marco},
	booktitle = {Proceedings of the 3rd Special Track on Big Data and High-Performance Computing (BigHPC 2025) co-located with the 4th Italian Conference on Big Data and Data Science (ITADATA 2025), Turin, Italy, September 9-11, 2025.},
	issn = {1613-0073},
	keywords = {icsc, space, hpc, fl},
	month = sep,
	pages = {1--5},
	publisher = {{CEUR}},
	title = {Benchmarking Federated Learning Frameworks' Scalability},
	url = {https://hdl.handle.net/2318/1933852},
	urn = {urn:nbn:de:0074-4124-x},
	volume = {4124},
	year = {2025},
	bdsk-url-1 = {https://hdl.handle.net/2318/1933852}}

@inproceedings{Malenza_P3HPC_24,
	abstract = {Applications that analyze data from modern scientific experiments will soon require a computing capacity of ExaFLOPs. The current trend to achieve such performance is to employ GPU-accelerated supercomputers and design applications to optimally exploit this hardware. Since each supercomputer is typically a one-off project, the necessity of having computational languages portable across diverse CPU and GPU architectures without performance losses is increasingly compelling. Here, we study the performance portability of the LSQR algorithm as found in the AVU-GSR code of the ESA Gaia mission. This code computes the astrometric parameters of the ~108 stars in our Galaxy. The LSQR algorithm is widely used across a broad range of high-performance computing (HPC) applications, elevating the study's relevance beyond the astrophysical domain. We developed different GPU-accelerated ports based on CUDA, C++ PSTL, SYCL, OpenMP, and HIP. We carefully verified the correctness of each port and tuned them to five different GPU-accelerated platforms from NVIDIA and AMD to evaluate the performance portability (PP) in terms of the harmonic mean of the application's performance efficiency across the tested hardware. HIP was demonstrated to be the most portable solution with a 0.94 average PP across the tested problem sizes, closely followed by SYCL coupled with AdaptiveCpp (ACPP) with 0.93. If we only consider NVIDIA platforms, CUDA would be the winner with 0.97. The tuning-oblivious C++ PSTL achieves 0.62 when coupled with vendor-specific compilers.},
	author = {Malenza, Giulio and Cesare, Valentina and Santimaria, Marco Edoardo and Birke, Robert and Vecchiato, Alberto and Becciani, Ugo and Aldinucci, Marco},
	booktitle = {SC24-W: Workshops of the International Conference for High Performance Computing, Networking, Storage and Analysis},
	date-modified = {2025-05-24 09:39:12 +0200},
	doi = {10.1109/SCW63240.2024.00157},
	isbn = {979-8-3503-5554-3},
	keywords = {astro, icsc, eupex},
	pages = {1152--1163},
	publisher = {IEEE},
	title = {Performance portability via C++ PSTL, SYCL, OpenMP, and HIP: the Gaia AVU-GSR case study},
	url = {https://conferences.computer.org/sc-wpub/pdfs/SC-W2024-6oZmigAQfgJ1GhPL0yE3pS/555400b152/555400b152.pdf},
	year = {2024},
	bdsk-url-1 = {https://conferences.computer.org/sc-wpub/pdfs/SC-W2024-6oZmigAQfgJ1GhPL0yE3pS/555400b152/555400b152.pdf}}

@inproceedings{24:garcia:itadata,
	abstract = {The rising usage of compute-intensive AI applications with fast response time requirements, such as text generation using large language models, underscores the need for more efficient and versatile hardware solutions. This drives the exploration of emerging architectures like RISC-V, which has the potential to deliver strong performance within tight power constraints. The recent commercial release of processors with RISC-V Vector (RVV) silicon-enabled extensions further amplifies the significance of RISC-V architectures, offering enhanced capabilities for parallel processing and accelerating tasks critical to large language models and other AI applications. This work aims to evaluate the BERT and GPT-2 language models inference performance on the SOPHON SG2042 64-core RISC-V architecture with silicon-enabled RVV v0.7.1. We benchmarked the models with and without RVV, using OpenBLAS and BLIS as BLAS backends for PyTorch to enable vectorization. Enabling RVV in OpenBLAS improved the inference performance by up to 40\% in some cases.},
	author = {Garcia, Adriano Marques and Malenza, Giulio and Birke, Robert and Aldinucci, Marco},
	booktitle = {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},
	editor = {Alessia Antelmi and Emanuele Carlini and Patrizio Dazzi},
	keywords = {eupilot, icsc},
	location = {Pisa, Italy},
	publisher = {CEUR-WS.org},
	series = {{CEUR} Workshop Proceedings},
	title = {{Assessing Large Language Models Inference Performance on a 64-core RISC-V CPU with Silicon-Enabled Vectors}},
	url = {https://iris.unito.it/retrieve/1540f675-5e88-4f57-95e7-df8e0fe5f1df/paper110.pdf},
	volume = {3785},
	year = {2024},
	bdsk-url-1 = {https://iris.unito.it/retrieve/1540f675-5e88-4f57-95e7-df8e0fe5f1df/paper110.pdf}}

@article{24:eurohpc:xffl,
	abstract = {In a decade, AI frontier research transitioned from the researcher's workstation to thousands of high-end hardware-accelerated compute nodes. This rapid evolution shows no signs of slowing down in the foreseeable future. While top cloud providers may be able to keep pace with this growth rate, obtaining and efficiently exploiting computing resources at that scale is a daunting challenge for universities and SMEs. This work introduces the Cross-Facility Federated Learning (XFFL) framework to bridge this compute divide, extending the opportunity to efficiently exploit multiple independent data centres for extreme-scale deep learning tasks to data scientists and domain experts. XFFL relies on hybrid workflow abstractions to decouple tasks from environment-specific technicalities, reducing complexity and enhancing reusability. In addition, Federated Learning (FL) algorithms eliminate the need to move large amounts of data between different facilities, reducing time-to-solution and preserving data privacy. The XFFL approach is empirically evaluated by training a full LLaMAv2 7B instance on two facilities of the EuroHPC JU, showing how the increased computing power completely compensates for the additional overhead introduced by two data centres.},
	address = {Bruxelles, Belgium},
	author = {Colonnelli, Iacopo and Birke, Robert and Malenza, Giulio and Mittone, Gianluca and Mulone, Alberto and Galjaard, Jeroen and Chen, Lydia Y. and Bassini, Sanzio and Scipione, Gabriella and Martinovi\v{c}, Jan and Vondr\'{a}k, Vit and Aldinucci, Marco},
	booktitle = {Proceedings of the First EuroHPC user day},
	date-modified = {2025-09-23 17:53:31 +0200},
	doi = {10.1016/j.procs.2024.07.003},
	issn = {1877-0509},
	journal = {Procedia Computer Science},
	keywords = {streamflow, space, icsc, fl, hpc},
	pages = {3--12},
	publisher = {Elsevier},
	title = {Cross-Facility Federated Learning},
	url = {https://www.sciencedirect.com/science/article/pii/S1877050924016909},
	volume = {240},
	year = {2024},
	bdsk-url-1 = {https://www.sciencedirect.com/science/article/pii/S1877050924016909},
	bdsk-url-2 = {https://doi.org/10.1016/j.procs.2024.07.003}}

@inproceedings{24:queyrut:srds,
	author = {Simon Queyrut and Robert Birke and Pascal Felber and Valerio Schiavon},
	booktitle = {43rd International Symposium on Reliable Distributed Systems {SRDS}},
	date-modified = {2024-07-28 00:23:59 +0200},
	keywords = {icsc, ai},
	title = {CLUES: Collusive Theft of Conditional Generative Adversarial Networks},
	year = {2024}}

@inproceedings{24:betianu:llm,
	abstract = {Large language models (LLMs) increasingly serve as the backbone for classifying text associated with distinct domains and simultaneously several labels (classes). When encountering domain shifts, e.g., classifier of movie reviews from IMDb to Rotten Tomatoes, adapting such an LLM-based multi-label classifier is challenging due to incomplete label sets at the target domain and daunting training overhead. The existing domain adaptation methods address either image multi-label classifiers or text binary classifiers. In this paper, we design DALLMi, Domain Adaptation Large Language Model interpolator, a first-of-its-kind semi-supervised domain adaptation method for text data models based on LLMs, specifically BERT. The core of DALLMi is the novel variation loss and MixUp regularization, which jointly leverage the limited positively labeled and large quantity of unlabeled text and, importantly, their interpolation from the BERT word embeddings. DALLMi also introduces a label-balanced sampling strategy to overcome the imbalance between labeled and unlabeled data. We evaluate DALLMi against the partial-supervised and unsupervised approach on three datasets under different scenarios of label availability for the target domain. Our results show that DALLMi achieves higher mAP than unsupervised and partially-supervised approaches by 19.9\% and 52.2\%, respectively.},
	author = {Betianu, Miruna and Malan, Abele and Aldinucci, Marco and Birke, Robert and Chen, Lydia Y.},
	booktitle = {Proceedings of the 28th Pacific-Asia Conference on Knowledge Discovery and Data Mining},
	date-modified = {2025-05-24 10:34:46 +0200},
	doi = {10.1007/978-981-97-2259-4_21},
	editor = {De{-}Nian Yang and Xing Xie and Vincent S. Tseng and Jian Pei and Jen{-}Wei Huang and Jerry Chun{-}Wei Lin},
	keywords = {icsc, eupilot, ai},
	location = {Taipei, Taiwan},
	month = may,
	pages = {277--289},
	publisher = {Springer},
	series = {Lecture Notes in Computer Science},
	title = {{DALLMi}: Domain Adaption for LLM-Based Multi-label Classifier},
	url = {https://hdl.handle.net/2318/1976672},
	volume = {14647},
	year = {2024},
	bdsk-url-1 = {https://hdl.handle.net/2318/1976672},
	bdsk-url-2 = {https://doi.org/10.1007/978-981-97-2259-4%5C_21}}

@inproceedings{24:chen:llm,
	abstract = {Knowledge distillation has been applied on generative models, such as Variational Autoencoder (VAE) and Generative Adversarial Networks (GANs). To distill the knowledge, the synthetic outputs of a teacher generator are used to train a student model. While the dark knowledge, i.e., the probabilistic output, is well explored in distilling classifiers, little is known about the existence of an equivalent dark knowledge for generative models and its extractability. In this paper, we derive the first kind of empirical risk bound for distilling generative models from a Bayesian perspective. Through our analysis, we show the existence of the dark knowledge for generative models, i.e., Bayes probability distribution of a synthetic output from a given input, which achieves lower empirical risk bound than merely using the synthetic output of the generators. Furthermore, we propose a {D}ark {K}nowledge based {D}istillation , {DKtill}, which trains the student generator based on the (approximate) dark knowledge. Our extensive evaluation on distilling VAE, conditional GANs, and translation GANs on Facades and CelebA datasets show that the FID of student generators trained by {DKtill} combining dark knowledge are lower than student generators trained only by the synthetic outputs by up to 42.66\%, and 78.99\%, respectively.},
	author = {Hong, Chi and Birke, Robert and Chen, Pin-Yu and Chen, Lydia Y.},
	booktitle = {Proceedings of the 28th Pacific-Asia Conference on Knowledge Discovery and Data Mining},
	doi = {10.1007/978-981-97-2253-2\_19},
	editor = {De{-}Nian Yang and Xing Xie and Vincent S. Tseng and Jian Pei and Jen{-}Wei Huang and Jerry Chun{-}Wei Lin},
	keywords = {icsc, epi, ai},
	location = {Taipei, Taiwan},
	month = may,
	pages = {235--247},
	publisher = {Springer},
	series = {Lecture Notes in Computer Science},
	title = {{On Dark Knowledge for Distilling Generators}},
	url = {https://hdl.handle.net/2318/1976671},
	volume = {14646},
	year = {2024},
	bdsk-url-1 = {https://hdl.handle.net/2318/1976671},
	bdsk-url-2 = {https://doi.org/10.1007/978-981-97-2253-2%5C_19}}

@inproceedings{24:casella:sgx,
	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, it becomes necessary to take additional security measures at the cost of increasing runtimes. The Trusted Execution Environment (TEE) approach promises to offer 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.},
	author = {Casella, Bruno and Colonnelli, Iacopo and Mittone, Gianluca and Birke, Robert and Riviera, Walter and Sciarappa, Antonio and Cavazzoni, Carlo and Aldinucci, Marco},
	booktitle = {Proceedings of the 2024 Deep Learning Security and Privacy Workshop, IEEE Symposium on Security and Privacy 2024},
	date-modified = {2024-08-18 13:16:08 +0200},
	doi = {10.1109/SPW63631.2024.00009},
	keywords = {icsc, epi, confidential, fl},
	location = {San Francisco, CA},
	month = may,
	title = {A Performance Analysis for Confidential Federated Learning},
	url = {https://iris.unito.it/retrieve/b5877a97-2d8d-4e95-8791-0aa4a1b953b3/DLSP___CONFIDENTIAL_FL.pdf},
	year = {2024},
	bdsk-url-1 = {https://iris.unito.it/retrieve/b5877a97-2d8d-4e95-8791-0aa4a1b953b3/DLSP___CONFIDENTIAL_FL.pdf}}

@inproceedings{23:mittone: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},
	author = {Gianluca Mittone and Giulio Malenza and Marco Aldinucci and Robert Birke},
	booktitle = {Proc. of the 16th IEEE/ACM Intl. Conference on Utility and Cloud Computing Companion {(UCC)}},
	date-modified = {2024-08-18 15:14:21 +0200},
	doi = {10.1145/3603166.3632561},
	institution = {Computer Science Department, University of Torino},
	keywords = {icsc, eupilot, fl},
	month = dec,
	note = {eupilot, icsc, fl},
	pages = {1--6},
	publisher = {{ACM}},
	title = {Distributed Edge Inference: an Experimental Study on Multiview Detection},
	url = {https://iris.unito.it/handle/2318/1950083},
	volume = {30},
	year = {2024},
	bdsk-url-1 = {https://iris.unito.it/handle/2318/1950083}}

@inproceedings{23:mittone:fl,
	abstract = {Federated Learning (FL) is experiencing a substantial research interest, with many frameworks being developed to allow practitioners to build federations easily and quickly. Most of these efforts do not consider two main aspects that are key to Machine Learning (ML) software: customizability and performance. This research addresses these issues by implementing an open-source FL framework named FastFederatedLearning (FFL). FFL is implemented in C/C++, focusing on code performance, and allows the user to specify any communication graph between clients and servers involved in the federation, ensuring customizability. FFL is tested against Intel OpenFL, achieving consistent speedups over different computational platforms (x86-64, ARM-v8, RISC-V), ranging from 2.5x and 3.69x. We aim to wrap FFL with a Python interface to ease its use and implement a middleware for different communication backends to be used. We aim to build dynamic federations in which relations between clients and servers are not static, giving life to an environment where federations can be seen as long-time evolving structures and exploited as services.},
	address = {Limassol, Cyprus},
	author = {Mittone, Gianluca and Birke, Robert and Aldinucci, Marco},
	booktitle = {Euro-Par 2023: Parallel Processing Workshops},
	doi = {10.1007/978-3-031-48803-0_40},
	institution = {Computer Science Department, University of Torino},
	keywords = {eupilot, icsc, hpc, fl},
	month = apr,
	pages = {321--326},
	publisher = {{Springer}},
	title = {Efficiently Distributed Federated Learning},
	url = {https://hdl.handle.net/2318/2031766},
	volume = {14352},
	year = {2024},
	bdsk-url-1 = {https://doi.org/10.1007/978-3-031-48803-0_40}}

@article{24:fdata:zhao,
	author = {Zhao, Zilong and Kunar, Aditya and Birke, Robert and Van der Scheer, Hiek and Chen, Lydia Y.},
	doi = {10.3389/FDATA.2023.1296508},
	journal = {Frontiers Big Data},
	keywords = {ai},
	title = {{CTAB-GAN+:} enhancing tabular data synthesis},
	url = {https://doi.org/10.3389/fdata.2023.1296508},
	volume = {6},
	year = {2024},
	bdsk-url-1 = {https://iris.unito.it/retrieve/42e86324-b7bd-4236-8458-4dfc24bbff76/fdata-06-1296508.pdf}}

@article{24:tnsm:nur,
	author = {Nur Zincir{-}Heywood and Robert Birke and Elias Bou{-}Harb and Takeru Inoue and Neeraj Kumar and Hanan Lutfiyya and Deepak Puthal and Abdallah Shami and Natalia Stakhanova},
	doi = {10.1109/TNSM.2024.3416861},
	journal = {{IEEE} Trans. Netw. Serv. Manag.},
	keywords = {ai},
	number = {3},
	pages = {2608--2612},
	title = {Guest Editorial: Special section on Networks, Systems, and Services Operations and Management Through Intelligence},
	url = {https://doi.org/10.1109/TNSM.2024.3416861},
	volume = {21},
	year = {2024},
	bdsk-url-1 = {https://ieeexplore.ieee.org/document/10599564/}}

@inproceedings{23:cloud:gdts,
	abstract = {Generative Adversarial Networks (GANs) are typically trained to synthesize data, from images and more recently tabular data, under the assumption of directly accessible training data. While learning image GANs on Federated Learning (FL) and Multi-Discriminator (MD) systems has just been demonstrated, it is unknown if tabular GANs can be learned from decentralized data sources. Different from image GANs, state-of-the-art tabular GANs require prior knowledge on the data distribution of each (discrete and continuous) column to agree on a common encoding -- risking privacy guarantees. In this paper, we propose GDTS, a distributed framework for GAN-based tabular synthesizer. GDTS provides different system architectures to match the two training paradigms termed GDTS FL and GDTS MD. Key to enable learning on distributed data is the proposed novel privacy-preserving multi-source feature encoding to capture the global data properties. In addition GDTS encompasses a weighting strategy based on table similarity to counter the detrimental effects of non-IID data and a validation pipeline to easily assess and compare the performance of different paradigms and hyper parameters. We evaluate the effectiveness of GDTS in terms of synthetic data quality, and overall training scalability. Experiments show that GDTS FL achieves better statistical similarity and machine learning utility between generated and original data compared to GDTS MD.},
	address = {Chicago, USA},
	author = {Zhao, Zilong and Birke, Robert and Chen, Lydia Y.},
	booktitle = {16th {IEEE} International Conference on Cloud Computing ({CLOUD})},
	doi = {10.1109/CLOUD60044.2023.00078},
	keywords = {, ai},
	month = jul,
	publisher = {{IEEE}},
	title = {{GDTS: GAN-based Distributed Tabular Synthesizer}},
	url = {https://iris.unito.it/retrieve/8bc610de-3ccd-4a0a-b97f-ee329e487b76/GDTS_IEEE_CLOUD_preprint.pdf},
	year = {2023},
	bdsk-url-1 = {https://iris.unito.it/retrieve/8bc610de-3ccd-4a0a-b97f-ee329e487b76/GDTS_IEEE_CLOUD_preprint.pdf},
	bdsk-url-2 = {https://doi.org/10.1109/CLOUD60044.2023.00078}}

@inproceedings{23:zhao:fctgan,
	abstract = {An alternative method for sharing knowledge while complying with strict data access regulations, such as the European General Data Protection Regulation (GDPR), is the emergence of synthetic tabular data. Mainstream table synthesizers utilize methodologies derived from Generative Adversarial Networks (GAN). Although several state-of-the-art (SOTA) tabular GAN algorithms inherit Convolutional Neural Network (CNN)-based architectures, which have proven effective for images, they tend to overlook two critical properties of tabular data: (i) the global correlation across columns, and (ii) the semantic invariance to the column order. Permuting columns in a table does not alter the semantic meaning of the data, but features extracted by CNNs can change significantly due to their limited convolution filter kernel size. To address the above problems, we propose FCT-GAN-- the first conditional tabular GAN to adopt Fourier networks into table synthesis. FCT-GAN enhances permutation invariant GAN training by strengthening the learning of global correlations via Fourier layers. Extensive evaluation on benchmarks and real-world datasets show that FCT-GAN can synthesize tabular data with better (up to 27.8%) machine learning utility (i.e. a proxy of global correlations) and higher (up to 26.5%) statistical similarity to real data. FCT-GAN also has the least variation on synthetic data quality among 7 SOTA baselines on 3 different training-data column orders.},
	address = {Birmingham, United Kingdom},
	author = {Zhao, Zilong and Birke, Robert and Chen, Lydia Y.},
	booktitle = {32nd {ACM} International Conference on Information and Knowledge Management ({CIKM '23})},
	doi = {10.1145/3583780.3615202},
	keywords = {icsc},
	month = oct,
	publisher = {{ACM}},
	title = {{FCT-GAN: Enhancing Global Correlation of Table Synthesis via Fourier Transform}},
	url = {https://iris.unito.it/retrieve/966ba767-dbbd-41e1-b4e3-7ab7ba09303f/FCT-GAN.pdf},
	year = {2023},
	bdsk-url-1 = {https://iris.unito.it/retrieve/966ba767-dbbd-41e1-b4e3-7ab7ba09303f/FCT-GAN.pdf},
	bdsk-url-2 = {https://doi.org/10.1145/3583780.3615202}}

@inproceedings{23:risc-v-summit,
	abstract = {RISC-V is an emerging instruction set architecture. Its modular and extensible open-source royalty-free design is increasingly attracting interest from both research and industry. Nowadays, different RISC-V-based boards can be bought off the shelf. However, software availability is equivalently vital in guaranteeing the RISC-V ecosystem's success. Here we contribute with the first publicly available port of PyTorch. PyTorch is one of the most popular Deep Learning libraries available today. As such, it is a crucial enabler in running state-of-the-art AI applications on RISC-V-based systems and a first step towards a fully democratic end-to-end codesign process.},
	address = {Barcelona, Spain},
	author = {Iacopo Colonnelli and Robert Birke and Marco Aldinucci},
	booktitle = {{RISC-V Summit Europe 2023}},
	keywords = {eupilot, icsc, riscv},
	month = jun,
	note = {Poster},
	title = {Experimenting with {PyTorch} on {RISC-V}},
	url = {https://iris.unito.it/retrieve/429bf344-9090-42c3-809c-1b8ac320a930/2023-06-08-Iacopo-COLONNELLI-abstract.pdf},
	year = {2023},
	bdsk-url-1 = {https://iris.unito.it/retrieve/429bf344-9090-42c3-809c-1b8ac320a930/2023-06-08-Iacopo-COLONNELLI-abstract.pdf}}

@inproceedings{23:SAMOS,
	abstract = {High-PerformanceComputing(HPC)haveevolvedtobeused to perform simulations of systems where physical experimentation is pro- hibitively impractical, expensive, or dangerous. This paper provides a general overview and showcases the analysis of non-functional properties in RISC-V-based platforms for HPCs. In particular, our analyses target the evaluation of power and energy control, thermal management, and reliability assessment of promising systems, structures, and technologies devised for current and future generation of HPC machines. The main set of design methodologies and technologies developed within the activ- ities of the Future and HPC & Big Data spoke of the National Centre of HPC, Big Data and Quantum Computing project are described along with the description of the testbed for experimenting two-phase cooling approaches.},
	address = {Samos, Greece},
	author = {Fornaciari, William and Reghenzani, Federico and Terraneo, Federico and Baroffio, Davide and Metra, Cecilia and Omana, Martin and Rodriguez Condia, Josie E. and Sonza Reorda, Matteo and Birke, Robert and Colonnelli, Iacopo and Mittone, Gianluca and Aldinucci, Marco and Mencagli, Gabriele and Iannone, Francesco and Palombi, Filippo and Zummo, Giuseppe and Cesarini, Daniele and Tesser, Federico},
	booktitle = {{Embedded Computer Systems: Architectures, Modeling, and Simulation - 23rd International Conference, {SAMOS} 2023}},
	doi = {10.1007/978-3-031-46077-7_26},
	keywords = {icsc, riscv, hpc},
	note = {icsc},
	title = {{RISC-V}-based Platforms for {HPC}: Analyzing Non-functional Properties for Future {HPC} and {Big-Data} Clusters},
	url = {https://iris.unito.it/retrieve/b627eab0-3aa1-4fd7-8685-f47c62c792b3/SAMOS_2023_CN_HPC_FL1.pdf},
	year = {2023},
	bdsk-url-1 = {https://iris.unito.it/retrieve/b627eab0-3aa1-4fd7-8685-f47c62c792b3/SAMOS_2023_CN_HPC_FL1.pdf}}

@inproceedings{23:aldinucci:continuum,
	abstract = {This paper proposes a continuum-aware programming model enabling the execution of application workflows across the compute continuum: cloud, fog and edge resources. It simplifies the management of heterogeneous nodes while alleviating the burden of programmers and unleashing innovation. This model optimizes the continuum through advanced development experiences by transforming workflows into autonomous service collaborations. It reduces complexity in positioning/interconnecting services across the continuum. A meta-model introduces high-level workflow descriptions as service networks with defined contracts and quality of service, thus enabling the deployment/management of workflows as first-class entities. It also provides automation based on policies, monitoring and heuristics. Tailored mechanisms orchestrate/manage services across the continuum, optimizing performance, cost, data protection and sustainability while managing risks. This model facilitates incremental development with visibility of design impacts and seamless evolution of applications and infrastructures. In this work, we explore this new computing paradigm showing how it can trigger the development of a new generation of tools to support the compute continuum progress.},
	address = {Turin, Italy},
	author = {Aldinucci, Marco and Birke, Robert and Brogi, Antonio and Carlini, Emanuele and Coppola, Massimo and Danelutto, Marco and Dazzi, Patrizio and Ferrucci, Luca and Forti Stefano and Kavalionak, Hanna and Mencagli, Gabriele and Mordacchin, Matteo and Pasin, Marcelo and Paganelli, Federica and Torquati, Massimo},
	booktitle = {{2023 IEEE 47th Annual Computers, Software, and Applications Conference (COMPSAC)}},
	doi = {10.1109/COMPSAC57700.2023.00287},
	keywords = {icsc},
	month = jun,
	publisher = {{IEEE}},
	title = {{A Proposal for a Continuum-aware Programming Model: From Workflows to Services Autonomously Interacting in the Compute Continuum}},
	url = {https://iris.unito.it/retrieve/2ae13a33-5814-43da-8ea6-2d3e8b122384/Continuum-aware-PM.pdf},
	year = {2023},
	bdsk-url-1 = {https://iris.unito.it/retrieve/2ae13a33-5814-43da-8ea6-2d3e8b122384/Continuum-aware-PM.pdf},
	bdsk-url-2 = {https://doi.org/10.1109/COMPSAC57700.2023.00287}}

@inproceedings{23:hong:datafree,
	abstract = {Deep machine learning models, e.g., image classifier, are increasingly deployed in the wild to provide services to users. Adversaries are shown capable of stealing the knowledge of these models by sending inference queries and then training substitute models based on query results. The availability and quality of adversarial query inputs are undoubtedly crucial in the stealing process. The recent prior art demonstrates the feasibility of replacing real data by exploring the synthetic adversarial queries, so called data-free attacks, under strong adversarial assumptions, i.e., the deployed classier returns not only class labels but also class probabilities. In this paper, we consider a general adversarial model and propose an effective data-free stealing algorithm, Tandem-GAN, which not only explores synthetic queries but also explicitly exploits the high quality ones. The core of TandemGAN is composed of (i) substitute model which imitates the target model through synthetic queries and their inferred labels; and (ii) a tandem generator consisting of two networks, Gx and Ge, which first explores the synthetic data space via Gx and then exploits high-quality examples via Ge to maximize the knowledge transfer from the target to the substitute model. Our results on four datasets show that the accuracy of our trained substitute model ranges between 96-67\% of the target model and outperforms the existing state-of-the-art data-free model stealing approach by up to 2.5X.},
	address = {Turin, Italy},
	author = {Hong, Chi and Huang, Jiyue and Birke, Robert and Chen, Lydia Y.},
	booktitle = {{European Conference on Machine Learning and Principles and Practice of Knowledge Discovery in Databases (ECML PKDD)}},
	doi = {10.1007/978-3-031-43424-2_2},
	keywords = {icsc, eupilot},
	month = sep,
	title = {{Exploring and Exploiting Data-Free Model Stealing}},
	url = {https://iris.unito.it/retrieve/ce44dec6-12c9-443d-99e7-f1141e50aa3a/Data-free%20Model%20Stealing.pdf},
	year = {2023},
	bdsk-url-1 = {https://dx.doi.org/10.1007/978-3-031-43424-2_2}}

@inproceedings{23:valtari:pilot,
	abstract = {Virtualized protection and control (VPC) is seen as a promising evolution for the centralized protection and control (CPC) concept. Centralization of protection functions consolidates the functions of multiple traditional relays into one device. This consolidation reduces communications network complexity and offers effective ways to manage protection applications of the substation. Making the CPC available as a VPC software image instead of a dedicated device creates yet another degree of freedom. The solution becomes hardware independent, bringing more flexibility and scalability to the solution. ABB and Caruna together wanted to explore these possibilities in a real-life substation pilot. This paper describes the piloted VPC environment and the results from the piloting period. The results show that virtualization technology is suitable for time critical protection and control applications, with real-time performance comparable to existing non- virtualized solutions.},
	address = {Rome, Italy},
	author = {Valtari, Jani and Kulmala, Anna and Sch\"{o}nborn, Sandro and Khozaya, David and Birke, Robert and Reikko Jyrki},
	booktitle = {{27th International Conference on Electricity Distribution (CIRED)}},
	date-modified = {2024-08-18 14:40:14 +0200},
	doi = {10.1049/icp.2023.1219},
	keywords = {RT},
	month = jun,
	title = {{Real-life Pilot of Virtual Protection and Control - Experiences and Performance Analysis}},
	url = {https://iris.unito.it/retrieve/5de5fb00-02bf-4ba8-a4db-5876415d5105/virtualization_full_paper_cired2023_submitted.pdf},
	year = {2023},
	bdsk-url-1 = {https://iris.unito.it/retrieve/5de5fb00-02bf-4ba8-a4db-5876415d5105/virtualization_full_paper_cired2023_submitted.pdf},
	bdsk-url-2 = {https://doi.org/10.1049/icp.2023.1219}}

@inproceedings{23:schoenborn:vipac,
	abstract = {Substation automation is ever challenged by the integration of distributed energy resources which imposes higher deployment flexibility and adaptability for protection and control. Although virtualization helps to run software applications independent of the underlying platform in IT infrastructures and cloud computing, it is still not commonly used in the field of substation automation. This is mainly due to the real-time performance demands of substation automation protection and control applications. In this article, we present an approach for running substation automation protection and control software in virtual environments. We contrast the real-time performance of different virtualization technologies under different workloads and focus on the performance evaluation of protection and control software in container- based solutions running on Linux with PREEMPT RT. We also present additional results for performance achieved in virtual machines. Our results clearly demonstrate that it is possible to run substation automation protection and control software in virtual environments while still providing the necessary performance. This paves the way for the deployment of substation protection and control software in virtualisation environments.},
	address = {Rome, Italy},
	author = {Sch\"{o}nborn, Sandro and Birke, Robert and Kozhaya, David and Sivanthi, Thanikesavan},
	booktitle = {{27th International Conference on Electricity Distribution (CIRED)}},
	date-modified = {2024-08-18 14:39:46 +0200},
	doi = {10.1049/icp.2023.1028},
	keywords = {RT},
	month = jun,
	title = {{Real-Time Performance of Virtualised Protection and Control Software}},
	url = {https://iris.unito.it/retrieve/eb610327-6e38-4f5e-8673-e62f2d956821/10702-Scho%cc%88nborn.pdf},
	year = {2023},
	bdsk-url-1 = {https://iris.unito.it/retrieve/eb610327-6e38-4f5e-8673-e62f2d956821/10702-Scho%cc%88nborn.pdf},
	bdsk-url-2 = {https://doi.org/10.1049/icp.2023.1028}}

@inproceedings{23:mittone:mafl,
	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 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. We test MAFL from multiple points of view, assessing its correctness, flexibility and scaling properties up to 64 nodes. We optimised the base software achieving a 5.5x speedup on a standard FL scenario. MAFL is compatible with x86-64, ARM-v8, Power and RISC-V.},
	address = {Limassol, Cyprus},
	author = {Mittone, Gianluca and Riviera, Walter and Colonnelli, Iacopo and Birke, Robert and Aldinucci, Marco},
	booktitle = {Euro-Par 2023: Parallel Processing},
	date-added = {2023-03-8 21:51:14 +0000},
	date-modified = {2024-08-18 13:15:50 +0200},
	doi = {10.1007/978-3-031-39698-4_26},
	institution = {Computer Science Department, University of Torino},
	keywords = {eupilot, icsc, riscv, fl},
	month = aug,
	pages = {383--396},
	publisher = {{Springer}},
	title = {Model-Agnostic Federated Learning},
	url = {https://hdl.handle.net/2318/1928750},
	volume = {14100},
	year = {2023},
	bdsk-url-1 = {https://doi.org/10.48550/arXiv.2303.04906}}

@inproceedings{23:mittone:fl-riscv,
	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 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 FastFlow parallel programming library. We experiment with it by generating different working DML schemes on two emerging architectures (ARM-v8, RISC-V) and the x86-64 platform. 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.},
	address = {Bologna, Italy},
	author = {Mittone, Gianluca and Tonci, Nicol{\`o} and Birke, Robert and Colonnelli, Iacopo and Medi\'{c}, Doriana and Bartolini, Andrea and Esposito, Roberto and Parisi, Emanuele and Beneventi, Francesco and Polato, Mirko and Torquati, Massimo and Benini, Luca and Aldinucci, Marco},
	booktitle = {20th {ACM} International Conference on Computing Frontiers ({CF} '23)},
	date-added = {2023-03-14 15:34:00 +0000},
	date-modified = {2024-08-18 13:15:37 +0200},
	doi = {10.1145/3587135.3592211},
	institution = {Computer Science Department, University of Torino},
	isbn = {979-8-4007-0140-5/23/05},
	keywords = {eupilot, icsc, hpc, riscv, fl},
	month = may,
	note = {https://arxiv.org/abs/2302.07946},
	publisher = {{ACM}},
	title = {Experimenting with Emerging {RISC-V} Systems for Decentralised Machine Learning},
	url = {https://dl.acm.org/doi/pdf/10.1145/3587135.3592211},
	year = {2023},
	bdsk-url-1 = {https://dl.acm.org/doi/pdf/10.1145/3587135.3592211},
	bdsk-url-2 = {https://doi.org/10.1145/3587135.3592211}}

@inproceedings{sdm-ghiassi23,
	abstract = {Learning robust deep models against noisy labels becomes ever critical when today's data is commonly collected from open platforms and subject to adversarial corruption. The information on the label corruption process, i.e., corruption matrix, can greatly enhance the robustness of deep models but still fall behind in combating hard classes. In this paper, we propose to construct a golden symmetric loss (GSL) based on the estimated corruption matrix as to avoid overfitting to noisy labels and learn effectively from hard classes. GSL is the weighted sum of the corrected regular cross entropy and reverse cross entropy. By leveraging a small fraction of trusted clean data, we estimate the corruption matrix and use it to correct the loss as well as to determine the weights of GSL. We theoretically prove the robustness of the proposed loss function in the presence of dirty labels. We provide a heuristics to adaptively tune the loss weights of GSL according to the noise rate and diversity measured from the dataset. We evaluate our proposed golden symmetric loss on both vision and natural language deep models subject to different types of label noise patterns. Empirical results show that GSL can significantly outperform the existing robust training methods on different noise patterns, showing accuracy improvement up to 18% on CIFAR-100 and 1% on real world noisy dataset of Clothing1M.},
	author = {Ghiassi, Amirmasoud and Birke, Robert and Chen, Lydia Y.},
	booktitle = {{SDM} '23: {SIAM} International Conference on Data Mining},
	doi = {10.1137/1.9781611977653.ch64},
	keywords = {textarossa},
	pages = {568--576},
	title = {Robust Learning via Golden Symmetric Loss of (un)Trusted Labels},
	url = {https://datacloud.di.unito.it/index.php/s/b6z3moNLxnNiCxz},
	year = {2023},
	bdsk-url-1 = {https://datacloud.di.unito.it/index.php/s/b6z3moNLxnNiCxz},
	bdsk-url-2 = {https://doi.org/10.1137/1.9781611977653.ch64}}

@article{COX2022101594,
	abstract = {Deep neural networks (DNNs) are becoming the core components of many applications running on edge devices, especially for real time image-based analysis. Increasingly, multi-faced knowledge is extracted by executing multiple DNNs inference models, e.g., identifying objects, faces, and genders from images. It is of paramount importance to guarantee low response times of such multi-DNN executions as it affects not only users quality of experience but also safety. The challenge, largely unaddressed by the state of the art, is how to overcome the memory limitation of edge devices without altering the DNN models. In this paper, we design and implement Masa, a responsive memory-aware multi-DNN execution and scheduling framework, which requires no modification of DNN models. The aim of Masa is to consistently ensure the average response time when deterministically and stochastically executing multiple DNN-based image analyses. The enabling features of Masa are (i) modeling inter- and intra-network dependency, (ii) leveraging complimentary memory usage of each layer, and (iii) exploring the context dependency of DNNs. We verify the correctness and scheduling optimality via mixed integer programming. We extensively evaluate two versions of Masa, context-oblivious and context-aware, on three configurations of Raspberry Pi and a large set of popular DNN models triggered by different generation patterns of images. Our evaluation results show that Masa can achieve lower average response times by up to 90% on devices with small memory, i.e., 512 MB to 1 GB, compared to the state of the art multi-DNN scheduling solutions.},
	author = {Cox, Bart and Birke, Robert and Chen, Lydia Y.},
	date-modified = {2024-04-01 11:32:22 +0200},
	doi = {https://doi.org/10.1016/j.pmcj.2022.101594},
	issn = {1574-1192},
	journal = {Pervasive and Mobile Computing},
	keywords = {ai},
	pages = {1--16},
	title = {Memory-aware and context-aware multi-DNN inference on the edge},
	url = {https://www.sciencedirect.com/science/article/pii/S1574119222000372},
	volume = {83},
	year = {2022},
	bdsk-url-1 = {https://www.sciencedirect.com/science/article/pii/S1574119222000372},
	bdsk-url-2 = {https://doi.org/10.1016/j.pmcj.2022.101594}}

@inproceedings{icpp-stewart22,
	abstract = {Short-term cache allocation grants and then revokes access to processor cache lines dynamically. For online services, short-term allocation can speed up targeted query executions and free up cache lines reserved, but normally not needed, for performance. However, in collocated settings, short-term allocation can increase cache contention, slowing down collocated query executions. To offset slowdowns, collocated services may request short-term allocation more often, making the problem worse. Short-term allocation policies manage which queries receive cache allocations and when. In collocated settings, these policies should balance targeted query speedups against slowdowns caused by recurring cache contention. We present a model-driven approach that (1) predicts response time under a given policy, (2) explores competing policies and (3) chooses policies that yield low response time for all collocated services. Our approach profiles cache usage offline, characterizes the effects of cache allocation policies using deep learning techniques and devises novel performance models for short-term allocation with online services. We tested our approach using data processing, cloud, and high-performance computing benchmarks collocated on Intel processors equipped with Cache Allocation Technology. Our models predicted median response time with 11\% absolute percent error. Short-term allocation policies found using our approach out performed state-of-the-art shared cache allocation policies by 1.2-2.3X.},
	author = {Stewart, Christopher and Morris, Nathaniel and Chen, Lydia Y. and Birke, Robert},
	booktitle = {Proceedings of the 51st International Conference on Parallel Processing ({ICPP})},
	date-modified = {2024-08-18 14:41:40 +0200},
	doi = {10.1145/3545008.3545094},
	keywords = {parallel},
	month = aug,
	pages = {31:1--31:11},
	publisher = {{ACM}},
	title = {Performance Modeling for Short-Term Cache Allocation},
	url = {https://doi.org/10.1145/3545008.3545094},
	year = {2022},
	bdsk-url-1 = {https://doi.org/10.1145/3545008.3545094}}

@inproceedings{bigdata-zhu22,
	abstract = {Tabular data synthesis is an emerging approach to circumvent strict regulations on data privacy while discovering knowledge through big data. Although state-of-the-art AI-based tabular data synthesizers, e.g., table-GAN, CTGAN, TVAE, and CTAB-GAN, are effective at generating synthetic tabular data, their training is sensitive to column permutations of input data. In this paper, we first c onduct a n e xtensive e mpirical s tudy to disclose such a property of permutation invariance and an in-depth analysis of the existing synthesizers. We show that changing the input column order worsens the statistical difference between real and synthetic data by up to 38.67\% due to the encoding of tabular data and the network architectures. To fully unleash the potential of big synthetic tabular data, we propose two solutions: (i) AE-GAN, a synthesizer that uses an autoencoder network to represent the tabular data and GAN networks to synthesize the latent representation, and (ii) a feature sorting algorithm to find t he s uitable c olumn o rder o f i nput d ata f or CNN-based synthesizers. We evaluate the proposed solutions on five datasets in terms of the sensitivity to the column permutation, the quality of synthetic data, and the utility in downstream analyses. Our results show that we enhance the property of permutation-invariance when training synthesizers and further improve the quality and utility of synthetic data, up to 22\%, compared to the existing synthesizers.},
	author = {Zhu, Yujin and Zhao, Zilong and Birke, Robert and Chen, Lydia Y.},
	booktitle = {{IEEE} International Conference on Big Data (Big Data)},
	date-modified = {2024-08-18 14:41:08 +0200},
	doi = {10.1109/BigData55660.2022.10020639},
	editor = {Shusaku Tsumoto and Yukio Ohsawa and Lei Chen and Dirk Van den Poel and Xiaohua Hu and Yoichi Motomura and Takuya Takagi and Lingfei Wu and Ying Xie and Akihiro Abe and Vijay Raghavan},
	keywords = {analytics},
	month = dec,
	pages = {5855--5864},
	publisher = {{IEEE}},
	title = {Permutation-Invariant Tabular Data Synthesis},
	url = {https://datacloud.di.unito.it/index.php/s/b6z3moNLxnNiCxz},
	year = {2022},
	bdsk-url-1 = {https://datacloud.di.unito.it/index.php/s/b6z3moNLxnNiCxz},
	bdsk-url-2 = {https://doi.org/10.1109/BigData55660.2022.10020639}}

@inproceedings{ghiassi/iccart22,
	abstract = {Today, to label the massive datasets needed to train Deep Neural Networks (DNNs), cheap and error-prone methods such as crowdsourcing are used. Label aggregation methods aim to infer the true labels from noisy labels annotated by crowdsourcing workers via labels statistics features. Aggregated labels are the main data source to train deep neural networks, and their accuracy directly affects the deep neural network performance. In this paper, we argue that training DNN and aggregating labels are not two separate tasks. Incorporation between DNN training and label aggregation connects data features, noisy labels, and aggregated labels. Since each image contains valuable knowledge about its label, the data features help aggregation methods enhance their performance. We propose LABNET an iterative two-step method. Step one: the label aggregation algorithm provides labels to train the DNN. Step two: the DNN shares a representation of the data features with the label aggregation algorithm. These steps are repeated until the converging label aggregation error rate. To evaluate LABNET we conduct an extensive empirical comparison on CIFAR-10 and CIFAR-100 under different noise and worker statistics. Our evaluation results show that LABNET achieves the highest mean accuracy with an increase of at least 8% to 0.6% and lowest error rate with a reduction of 7.5% to 0.25% against existing aggregation and training methods in most cases.},
	author = {Ghiassi, Amirmasoud and Birke, Robert and Chen, Lydia Y.},
	booktitle = {14th International Conference on Agents and Artificial Intelligence ({ICAART})},
	date-modified = {2025-05-24 09:18:24 +0200},
	doi = {10.5220/0010770400003116},
	editor = {Ana Paula Rocha and Luc Steels and H. Jaap van den Herik},
	keywords = {ai},
	month = feb,
	pages = {56--66},
	publisher = {{SCITEPRESS}},
	title = {{LABNET:} {A} Collaborative Method for {DNN} Training and Label Aggregation},
	url = {https://www.scitepress.org/Link.aspx?doi=10.5220/0010770400003116},
	year = {2022},
	bdsk-url-1 = {https://www.scitepress.org/Link.aspx?doi=10.5220/0010770400003116},
	bdsk-url-2 = {https://doi.org/10.5220/0010770400003116}}

@article{ZhaoBHRBMC21,
	abstract = {Classification algorithms have been widely adopted to detect anomalies for various systems, e.g., IoT, cloud and face recognition, under the common assumption that the data source is clean, i.e., features and labels are correctly set. However, data collected from the wild can be unreliable due to careless annotations or malicious data transformation for incorrect anomaly detection. In this article, we extend a two-layer on-line data selection framework: Robust Anomaly Detector (RAD) with a newly designed ensemble prediction where both layers contribute to the final anomaly detection decision. To adapt to the on-line nature of anomaly detection, we consider additional features of conflicting opinions of classifiers, repetitive cleaning, and oracle knowledge. We on-line learn from incoming data streams and continuously cleanse the data, so as to adapt to the increasing learning capacity from the larger accumulated data set. Moreover, we explore the concept of oracle learning that provides additional information of true labels for difficult data points. We specifically focus on three use cases, (i) detecting 10 classes of IoT attacks, (ii) predicting 4 classes of task failures of big data jobs, and (iii) recognising 100 celebrities faces. Our evaluation results show that RAD can robustly improve the accuracy of anomaly detection, to reach up to 98.95 percent for IoT device attacks (i.e., +7%), up to 85.03 percent for cloud task failures (i.e., +14%) under 40 percent label noise, and for its extension, it can reach up to 77.51 percent for face recognition (i.e., +39%) under 30 percent label noise. The proposed RAD and its extensions are general and can be applied to different anomaly detection algorithms.},
	author = {Zhao, Zilong and Birke, Robert and Han, Rui and Robu, Bogdan and Bouchenak, Sara and Ben Mokhtar, Sonia and Chen, Lydia Y.},
	date-modified = {2024-04-01 11:20:03 +0200},
	doi = {10.1109/TDSC.2021.3063947},
	journal = {{IEEE} Trans. Dependable Secur. Comput.},
	keywords = {ai},
	number = {5},
	pages = {2177--2192},
	title = {Enhancing Robustness of On-Line Learning Models on Highly Noisy Data},
	url = {https://doi.org/10.1109/TDSC.2021.3063947},
	volume = {18},
	year = {2021},
	bdsk-url-1 = {https://doi.org/10.1109/TDSC.2021.3063947}}

@article{BirkePQBC21,
	abstract = {Offering consistent low latency remains a key challenge for distributed applications, especially when deployed on the cloud where virtual machines (VMs) suffer from capacity variability caused by co-located tenants. Replicating redundant requests was shown to be an effective mechanism to defend application performance from high capacity variability. While the prior art centers on single-tier systems, it still remains an open question how to design replication strategies for distributed multi-tier systems. In this paper, we design a first of its kind PArtial REplication system, sPARE, that replicates and dispatches read-only workloads for distributed multi-tier web applications. The two key components of sPARE are (i) the variability-aware replicator that coordinates the replication levels on all tiers via an iterative searching algorithm, and (ii) the replication-aware arbiter that uses a novel token-based arbitration algorithm (TAD) to dispatch requests in each tier. We evaluate sPARE on web serving and searching applications, i.e., MediaWiki and Solr, the former deployed on our private cloud and the latter on Amazon EC2. Our results based on various interference patterns and traffic loads show that sPARE is able to improve the tail latency of MediaWiki and Solr by a factor of almost 2.7x and 2.9x, respectively.},
	author = {Birke, Robert and P{\'{e}}rez, Juan F. and Qiu, Zhan and Bj{"{o}}rkqvist, Mathias and Chen, Lydia Y.},
	date-modified = {2024-04-01 11:31:37 +0200},
	doi = {10.1109/TSC.2017.2780845},
	journal = {{IEEE} Trans. Serv. Comput.},
	keywords = {parallel},
	number = {2},
	pages = {574--588},
	title = {sPARE: Partial Replication for Multi-Tier Applications in the Cloud},
	url = {https://doi.org/10.1109/TSC.2017.2780845},
	volume = {14},
	year = {2021},
	bdsk-url-1 = {https://doi.org/10.1109/TSC.2017.2780845}}

@inproceedings{pmlr-v157-zhao21a,
	abstract = {While data sharing is crucial for knowledge development, privacy concerns and strict regulation (e.g., European General Data Protection Regulation (GDPR)) unfortunately limit its full effectiveness. Synthetic tabular data emerges as an alternative to enable data sharing while fulfilling regulatory and privacy constraints. The state-of-the-art tabular data synthesizers draw methodologies from Generative Adversarial Networks (GAN) and address two main data types in industry, i.e., continuous and categorical. In this paper, we develop CTAB-GAN, a novel conditional table GAN architecture that can effectively model diverse data types, including a mix of continuous and categorical variables. Moreover, we address data imbalance and long tail issues, i.e., certain variables have drastic frequency differences across large values. To achieve those aims, we first introduce the information loss, classification loss and generator loss to the conditional GAN. Secondly, we design a novel conditional vector, which efficiently encodes the mixed data type and skewed distribution of data variable. We extensively evaluate CTAB-GAN with the state of the art GANs that generate synthetic tables, in terms of data similarity and analysis utility. The results on five datasets show that the synthetic data of CTAB-GAN remarkably resembles the real data for all three types of variables and results into higher accuracy for five machine learning algorithms, by up to 17%.},
	author = {Zhao, Zilong and Kunar, Aditya and Birke, Robert and Chen, Lydia Y.},
	booktitle = {Proceedings of The 13th Asian Conference on Machine Learning},
	date-modified = {2025-05-24 09:21:39 +0200},
	editor = {Balasubramanian, Vineeth N. and Tsang, Ivor},
	keywords = {ai},
	month = nov,
	pages = {97--112},
	pdf = {https://proceedings.mlr.press/v157/zhao21a/zhao21a.pdf},
	publisher = {PMLR},
	series = {Proceedings of Machine Learning Research},
	title = {CTAB-GAN: Effective Table Data Synthesizing},
	url = {https://proceedings.mlr.press/v157/zhao21a.html},
	volume = {157},
	year = {2021},
	bdsk-url-1 = {https://proceedings.mlr.press/v157/zhao21a.html}}

@inproceedings{pmlr-v157-younesian21a,
	abstract = {Noisy labeled data is more a norm than a rarity for self-generated content that is continuously published on the web and social media from non-experts. Active querying experts are conventionally adopted to provide labels for the informative samples which don't have labels, instead of possibly incorrect labels. The new challenge that arises here is how to discern the informative and noisy labels which benefit from expert cleaning.  In this paper, we aim to leverage the stringent oracle budget to robustly maximize learning accuracy. We propose a noise-aware active learning framework, QActor, and a novel measure \emph{CENT}, which considers both cross-entropy and entropy to select informative and noisy labels for an expert cleansing. QActor iteratively cleans samples via quality models and actively querying an expert on those noisy yet informative samples. To adapt to learning capacity per iteration, QActor dynamically adjusts the query limit according to the learning loss for each learning iteration. We extensively evaluate different image datasets with noise label ratios ranging between 30% and 60%. Our results show that QActor can nearly match the optimal accuracy achieved using only clean data at the cost of only an additional 10% of ground truth data from the oracle.},
	author = {Younesian, Taraneh and Zhao, Zilong and Ghiassi, Amirmasoud and Birke, Robert and Chen, Lydia Y.},
	booktitle = {Proceedings of The 13th Asian Conference on Machine Learning},
	date-modified = {2025-05-24 09:19:25 +0200},
	editor = {Balasubramanian, Vineeth N. and Tsang, Ivor},
	keywords = {ai},
	month = nov,
	pages = {548--563},
	pdf = {https://proceedings.mlr.press/v157/younesian21a/younesian21a.pdf},
	publisher = {PMLR},
	series = {Proceedings of Machine Learning Research},
	title = {{QActor}: Active Learning on Noisy Labels},
	url = {https://proceedings.mlr.press/v157/younesian21a.html},
	volume = {157},
	year = {2021},
	bdsk-url-1 = {https://proceedings.mlr.press/v157/younesian21a.html}}

@inproceedings{bdcat-ghiassi21,
	abstract = {Big Data systems allow collecting massive datasets to feed the data hungry deep learning. Labelling these ever-bigger datasets is increasingly challenging and label errors affect even highly curated sets. This makes robustness to label noise a critical property for weakly-supervised classifiers. The related works on resilient deep networks tend to focus on a limited set of synthetic noise patterns, and with disparate views on their impacts, e.g., robustness against symmetric v.s. asymmetric noise patterns. In this paper, we first extend the theoretical analysis of test accuracy for any given noise patterns. Based on the insights, we design TrustNet that first learns the pattern of noise corruption, being it both symmetric or asymmetric, from a small set of trusted data. Then, TrustNet is trained via a robust loss function, which weights the given labels against the inferred labels from the learned noise pattern. The weight is adjusted based on model uncertainty across training epochs. We evaluate TrustNet on synthetic label noise for CIFAR-10, CIFAR-100 and big real-world data with label noise, i.e., Clothing1M. We compare against state-of-the-art methods demonstrating the strong robustness of TrustNet under a diverse set of noise patterns.},
	author = {Ghiassi, Amirmasoud and Birke, Robert and Chen, Lydia Y.},
	booktitle = {8th {IEEE/ACM} International Conference on Big Data Computing, Applications and Technologies ({BDCAT})},
	date-modified = {2025-05-24 09:18:42 +0200},
	doi = {10.1145/3492324.3494166},
	keywords = {ai},
	month = dec,
	pages = {52--62},
	publisher = {{ACM}},
	title = {{TrustNet}: Learning from Trusted Data Against (A)symmetric Label Noise},
	url = {https://doi.org/10.1145/3492324.3494166},
	year = {2021},
	bdsk-url-1 = {https://doi.org/10.1145/3492324.3494166}}

@inproceedings{etfa-albanese21,
	abstract = {Enterprises in the field of industrial automation experience an increasing demand for providing virtualized software solutions. Inspired by the recent trends in serverless and cloud computing, software virtualization is considered even for safety-critical applications with hard real-time requirements, as a means of avoiding hardware vendor lock-in and reducing volume and maintenance cost of devices. In this work, we evaluate the applicability of OS-level virtualization to an industrial automation use case. Our application runs in Docker containers on top of Linux patched with PREEMPT_RT. We investigate the ability of Docker coupled with diverse networking technologies to fulfill the latency requirements of the application under normal or heavy system load. We empirically compare four networking technologies with respect to communication latency and frequency of missing packets. The results indicate that Docker with certain technologies, such as the Single Root I/O Virtualization interface, performs robustly even under heavy load, enabling sufficient performance isolation and low overhead that does not jeopardise the real-time performance of our application.},
	author = {Giuliano Albanese and Robert Birke and Georgia Giannopoulou and Sandro Sch{\"{o}}nborn and Thanikesavan Sivanthi},
	booktitle = {26th {IEEE} International Conference on Emerging Technologies and Factory Automation ({ETFA})},
	date-modified = {2025-05-24 09:38:18 +0200},
	doi = {10.1109/ETFA45728.2021.9613320},
	editor = {Chiozzi, Gianluca and Ibsen, Jorge},
	month = sep,
	pages = {1--8},
	publisher = {{IEEE}},
	title = {Evaluation of Networking Options for Containerized Deployment of Real-Time Applications},
	url = {https://doi.org/10.1109/ETFA45728.2021.9613320},
	year = {2021},
	bdsk-url-1 = {https://doi.org/10.1109/ETFA45728.2021.9613320}}

@inproceedings{ijcnn-ghiassi21,
	abstract = {Today's available datasets in the wild, e.g., from social media and open platforms, present tremendous opportunities and challenges for deep learning, as there is a significant portion of tagged images, but often with noisy, i.e. erroneous, labels. Recent studies improve the robustness of deep models against noisy labels without the knowledge of true labels. In this paper, we advocate to derive a stronger classifier which proactively makes use of the noisy labels in addition to the original images - turning noisy labels into learning features. To such an end, we propose a novel framework, LABELNET, composed of Amateur and Expert, which iteratively learn from each other. Amateur is a regular image classifier trained by the feedback of Expert, which imitates how human experts would correct the predicted labels from Amateur using the noise pattern learnt from the knowledge of both the noisy and ground truth labels. The trained Amateur and Expert proactively leverage the images and their noisy labels to infer image classes. Our empirical evaluations on noisy versions of MNIST, CIFAR-10, CIFAR-100 and real-world data of Clothing1M show that the proposed model can achieve robust classification against a wide range of noise ratios and with as little as 20-50% training data, compared to state-of-the-art deep models that solely focus on distilling the impact of noisy labels.},
	author = {Ghiassi, Amirmasoud and Birke, Robert and Han, Rui and Chen, Lydia Y.},
	booktitle = {International Joint Conference on Neural Networks ({IJCNN})},
	date-modified = {2025-05-24 09:19:39 +0200},
	doi = {10.1109/IJCNN52387.2021.9533562},
	keywords = {ai},
	month = jul,
	pages = {1--8},
	publisher = {{IEEE}},
	title = {{LABELNET:} Recovering Noisy Labels},
	url = {https://doi.org/10.1109/IJCNN52387.2021.9533562},
	year = {2021},
	bdsk-url-1 = {https://doi.org/10.1109/IJCNN52387.2021.9533562}}

@inproceedings{percom-cox21a,
	abstract = {Deep neural networks (DNNs) are becoming the core components of many applications running on edge devices, especially for real time image-based analysis. Increasingly, multi-faced knowledge is extracted via executing multiple DNNs inference models, e.g., identifying objects, faces, and genders from images. The response times of multi-DNN highly affect users' quality of experience and safety as well. Different DNNs exhibit diversified resource requirements and execution patterns across layers and networks, which may easily exceed the available device memory and riskily degrade the responsiveness. In this paper, we design and implement Masa, a responsive memory-aware multi-DNN execution framework, an on-device middleware featuring on modeling inter- and intra-network dependency and leveraging complimentary memory usage of each layer. Masa can consistently ensure the average response time when deterministically and stochastically executing multiple DNN-based image analyses. We extensively evaluate Masa on three configurations of Raspberry Pi and a large set of popular DNN models triggered by different generation patterns of images. Our evaluation results show that Masa can achieve lower average response times by up to 90% on devices with small memory, i.e., 512 MB to 1 GB, compared to the state of the art multi-DNN scheduling solutions.},
	author = {Cox, Bart and Galjaard, Jeroen and Ghiassi, Amirmasoud and Birke, Robert and Chen, Lydia Y.},
	booktitle = {19th {IEEE} International Conference on Pervasive Computing and Communications ({PerCom})},
	doi = {10.1109/PERCOM50583.2021.9439111},
	month = mar,
	pages = {1--10},
	publisher = {{IEEE}},
	title = {Masa: Responsive Multi-DNN Inference on the Edge},
	url = {https://doi.org/10.1109/PERCOM50583.2021.9439111},
	year = {2021},
	bdsk-url-1 = {https://doi.org/10.1109/PERCOM50583.2021.9439111}}

@inproceedings{percom-galjaard21,
	abstract = {The execution of deep neural network (DNN) inference jobs on edge devices has become increasingly popular. Multiple of such inference models can concurrently analyse the on-device data, e.g. images, to extract valuable insights. Prior art focuses on low-power accelerators, compressed neural network architectures, and specialized frameworks to reduce execution time of single inference jobs on edge devices which are resource constrained. However, it is little known how different scheduling policies can further improve the runtime performance of multi-inference jobs without additional edge resources. To enable the exploration of scheduling policies, we first develop an execution framework, EdgeCaffe, which splits the DNN inference jobs by loading and execution of each network layer. We empirically characterize the impact of loading and scheduling policies on the execution time of multi-inference jobs and point out their dependency on the available memory space. We propose a novel memory-aware scheduling policy, MemA, which opportunistically interleaves the executions of different types of DNN layers based on their estimated run-time memory demands. Our evaluation on exhaustive combinations of five networks, data inputs, and memory configurations show that MemA can alleviate the degradation of execution times of multi-inference (up to 5x) under severely constrained memory compared to standard scheduling policies without affecting accuracy.},
	author = {Galjaard, Jeroen and Cox, Bart and Ghiassi, Amirmasoud and Chen, Lydia Y. and Birke, Robert},
	booktitle = {19th {IEEE} International Conference on Pervasive Computing and Communications Workshops and other Affiliated Events},
	date-modified = {2025-05-24 10:46:04 +0200},
	doi = {10.1109/PerComWorkshops51409.2021.9430952},
	keywords = {ai},
	month = mar,
	pages = {281--286},
	publisher = {{IEEE}},
	title = {{MemA}: Fast Inference of Multiple Deep Models},
	url = {https://doi.org/10.1109/PerComWorkshops51409.2021.9430952},
	year = {2021},
	bdsk-url-1 = {https://doi.org/10.1109/PerComWorkshops51409.2021.9430952}}

@inproceedings{www-hong21,
	abstract = {Noisy labeled data is more a norm than a rarity for crowd sourced contents. It is effective to distill noise and infer correct labels through aggregating results from crowd workers. To ensure the time relevance and overcome slow responses of workers, online label aggregation is increasingly requested, calling for solutions that can incrementally infer true label distribution via subsets of data items. In this paper, we propose a novel online label aggregation framework, BiLA , which employs variational Bayesian inference method and designs a novel stochastic optimization scheme for incremental training. BiLA is flexible to accommodate any generating distribution of labels by the exact computation of its posterior distribution. We also derive the convergence bound of the proposed optimizer. We compare BiLA with the state of the art based on minimax entropy, neural networks and expectation maximization algorithms, on synthetic and real-world data sets. Our evaluation results on various online scenarios show that BiLA can effectively infer the true labels, with an error rate reduction of at least 10 to 1.5 percent points for synthetic and real-world datasets, respectively.},
	author = {Chi Hong and Amirmasoud Ghiassi and Yichi Zhou and Robert Birke and Lydia Y. Chen},
	booktitle = {{WWW} '21: The Web Conference 2021},
	date-modified = {2024-04-01 11:20:18 +0200},
	doi = {10.1145/3442381.3449933},
	editor = {Jure Leskovec and Marko Grobelnik and Marc Najork and Jie Tang and Leila Zia},
	keywords = {ai},
	month = apr,
	pages = {1904--1915},
	publisher = {{ACM} / {IW3C2}},
	title = {Online Label Aggregation: {A} Variational Bayesian Approach},
	url = {https://doi.org/10.1145/3442381.3449933},
	year = {2021},
	bdsk-url-1 = {https://doi.org/10.1145/3442381.3449933}}

@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.}
}

@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}}

@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}}

@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}}

@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}

@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}}

@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}}

@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}}

@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}}

@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}}

@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}}

@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}}