Please download the program here
11:30
Monday 7 July 2025 – Day 1
11:30-12:00
Welcome and Registration: Aula Magna – Scuola Sant’Anna, Pisa
12:00-13:00
Networking Lunch
Edge AI and RISC-V Technology Advancements:
13:00-14:30
Plenary Session A01
13:00-14:30
Session A01 (Room Aula Magna – Scuola Sant’Anna): Edge and generative AI
13:00-14:30
Course A01-1: The path from edge fixed AI to GenAI: Tutorial and practical examples (Part 1 and 2). Danilo Pau, Technical Director, STMicroelectronics, Italy.
Course description:
Artificial Intelligence underwent to several key milestones through the years starting from the vision of Alan Turing, the perceptron of Rosenblatt, the existence theorem of Cybenko and Hornik and the seminal work of G. Hinton on Science. Next AlexNext by winning ILVRC on 2012 started the Fixed Function AI era during which more and more accurate fixed purpose AI workloads were developed and accelerated through GPUs. On 2019 TinyML community was established and the industry started to invest on new edge product culminating on frameworks and mass production low power micro controllers and neural processing units. This plethora of hardware, targeting various industry domains, including a variety of deep learning frameworks, neural network topologies, layers, development environment were the source of the HETEROGENOUS challenge which was a big obstacle to many stakeholders for the mass adoption of AI at the Edge. ST solved that challenge devising and delivering for free the Edge Unified AI Core Technology.
This technology interfaces the most widely used Deep Learning representation standards such as Google Keras, QKeras and Tensorflow Lite and the Open Neural Network Exchange (ONNX). It outputs optimized C code across heterogeneous instruction sets with public APIs for STM32, STM32N6, Stellar MCUs, and MEMs sensors. With that free of charge solution, ST helps the machine learning and embedded developer community to use the IDE they are comfortable with to finalize their application of choice with unprecedented execution speed. This tutorial includes practical demonstration of the tool in action on selected heterogenous hardware will be provided to the audience.
Lecturer(s): Danilo Pau, Technical Director, STMicroelectronics, Italy.
Danilo Pau: As of Dec 2024, I spent 32 years in STMicroelectronics and in System R&D. My (source GoogleScholar) h-index is 30, i10- index 86. I produced 109 invention’s requests, 80 EU and 71 US application patents. I am author of 224 scientific publications, 113 (2012-2015) ISO/IEC/MPEG documents and had 125 invited talks including key notes, seminars, tutorials at Universities/Conferences. I graduated on Electronic Engineering in 1992 at Politecnico di Milano after one year internship in SGS-THOMSON/CASA on 50% memory reduced HD-MAC HW decoder. I have a background on MPEG2 video memory reduction (SD/HD), video (up to H.264) encoding, transcoding, OpenVG/OpenGL-ES since Khronos was established, I co-chaired ISO/IEC SC29-WG11 CDVS and CDVA for 3 years, computer vision and tiny AI since 2016 (through the release of the 2024 Unified AI Core Technology) now part of STM32CubeMX, Stellar-Studio, SPC-Studio.AI, MEMs Studio, STM32 Developer Cloud and the Suite. On 2019 I was elevated IEEE Fellow; on 2022 AAIA Fellow, APSIPA life member since 2020, Sigma Xi honor society (invited) member since 2024. I served as Industry Ambassador (Italy section), coordinator IEEE Region 8 South Europe, vice-chairman of the “Intelligent Cyber-Physical Systems” Task Force IEEE CIS, IEEE R8 AfI internship initiative, AE IEEE TNNLS, Member of the IEEE CTSoc MDAI, 2023 IEEE CS Fellow EVM. I personally curated all the way the IEEE Milestone on Multiple Silicon Technologies on a chip, 1985 ratified by IEEE/2021, and IEEE Milestone on MPEG Multimedia Integrated Circuits, 1984-1993, ratified IEEE/2022, Integrated Circuits for Satellite Digital Radio, 1996-1997, ratified IEEE/2024. I serve TinyML (now EdgeAi) Foundation (Symposium, Summit, EMEA), chair of the TinyML on Device Learning, co-funded AutoTinyML, co-chair TinyML Talks. I am co-chair of EdgeAI Research track at EdgeAI Austin 2025 and co-chair of Edge GenAI working group. Moreover, coordinating contributions to MPAI Neural Watermarking IEEE stardardization. I was awarded with my team “Finmeccanica Innovation Award 2004 Embedded 3D graphics”, STAR Gold award 2014 for high performing Technical Staff (which I co-funded), best demo award with Politecnico di Milano on MPEG Visual Search, on 2014, STAR Silver award on 2019 for STM32Cube.AI, 2020 Spring APSIPA Industrial Distinguished Leaders Award, STAR Innovation Award in 2023. I supervised countless students over 30+ years. I strongly and openly cooperated with many professors, professionals and researchers and love continuing that. Innovation and be part of a scientific community are all my needs.
Lecture link:
14:30-16:00
Plenary Session B01
14:30-16:00
Session B01 (Room Aula Magna – Scuola Sant’Anna): Edge AI and autonomous systems
14:30-15:15
Course B01-1: Scenario generation for autonomous driving using LLMs based on real-world edge data. Andreas Stephan, Robert Bosch, Germany.
Course description:
This course explores scenario generation in the context of autonomous driving. It discusses the role of scenario generation in developing robust, and reliable autonomous driving systems. Participants will learn about common scenario representations and measures how to evaluate them. Furthermore, the course will give a high-level overview of traditional and state-of-the-art scenario generation techniques. These include rule-based methods, statistical methods, deep-learning based methods, and Large Language Model (LLM) driven approaches. The course culminates with an example architecture which leverages edge AI, LLMs and legal specifications to generate critical scenarios according to the provided specification.
Lecturer(s): Andreas Stephan, Robert Bosch, Germany.
Andreas, holds degrees in Mathematics and Computer Science from TU Munich and is a final year PhD candidate in Natural Language Processing (NLP) at the University of Vienna, supervised by Prof. Benjamin Roth. His research focuses on the integration and analysis of noisy sources of information in NLP. Prior to his academic engagement, he was in industry, tackling applied NLP challenges, including information extraction and the integration of graphs with textual data. Currently he works on scenario generation for ADAS systems at Bosch Research. He is broadly interested in NLP’s role in AI and its applications.
15:15-16:00
Course B01-2: Autonomous mobile robots in smart manufacturing: A real-world approach to edge AI and computer vision. Ayaan Verghese, Research Scientist, Dell Technologies, Ireland; Matthew Keating, Research Scientist, Dell Technologies, Ireland.
Course description:
The course explores the application of AI at the Edge and Photogrammetric Computer Vision for scene understanding to enable autonomous decision making in swarm networks. Currently large conventional factories, have static production lines that may remain unchanged for years. This lecture will demonstrate how centralized fixed manufacturing facilities can evolve using swarms of Autonomous Mobile Robots and smart edge devices communicating and collaborating at the edge, with minimal supervision by humans or central control systems. Students will gain insight into a low-code tool for training computer vision models, design of a reference architecture for edge deployments, the use of intelligent autonomous mobile robots in a manufacturing environment to help implement smart factories.
Lecturer(s): Ayaan Verghese, Research Scientist, Dell Technologies, Ireland; Matthew Keating, Research Scientist, Dell Technologies, Ireland.
Ayaan Verghese is a research scientist at Dell Technologies, exploring the potential of artificial intelligence, computer vision, and robotics to drive innovation and productivity in smart factories. With a strong background in manufacturing processes and industrial automation, he has previously been involved in the development and management of manufacturing processes for enterprise servers and storage systems, as well as deploying automation solutions in production environments. He currently collaborates with academic and industry partners on European Commission Horizon-funded projects, driving knowledge exchange and innovation. Holding a Master’s degree in Electrical and Electronics Engineering, he has published research on Quantum Computing for Electronic Design Automation. Ayaan brings a unique perspective on the intersection of technology and industry.
Matthew Keating is a Research Scientist at Dell Technologies’ Office of the CTO, where he explores the frontiers of Artificial Intelligence, Autonomous Systems, and Semantic Web technologies. With a strong background in industrial research, Matthew has collaborated with academia and industry partners on numerous European Commission Horizon-funded projects, driving innovation and research excellence in Europe. His expertise spans metadata management, machine learning, and earth observation, with a current focus on applying AI to satellite imagery for wildfire tracking and prediction. Matthew is nearing completion of his Master’s thesis on this topic. Matthew brings a unique blend of industry and academic experience, bringing a distinctive perspective to the intersection of technology and societal impact.
16:00-16:30
Afternoon Coffee/Tea Break
16:30-18:00
Plenary Session C01
16:30-18:00
Session C01 (Room Aula Magna – Scuola Sant’Anna): Edge AI deep neural networks (DNNs) tools
16:30-17:15
Course C01-1: Aidge – a new framework for DNN development, training and deployment on the edge. Fabian Chersi, Senior AI Architect, CEA-LIST, France.
Course description:
The aim of this course is to familiarize the participant with Aidge, a new comprehensive framework developed for simple and efficient DNN importing, construction, optimization, training, testing and deployment. The framework is particularly useful for DNN exploration and optimization, allowing simple and fast prototyping of DNNs with different topologies, quantization levels and optimization methods. It also allows to export to hardware targets such as CPUs, MCUs and GPUs as well as to custom hardware by means of High-Level Synthesis. The framework is developed in C++14 for high performance and provides python bindings for easiness and compatibility with major frameworks such as Pytorch, Tensorflow, ONNX runtime. The course comprises an introductory part and practical part.
Lecturer(s): Fabian Chersi, Senior AI Architect, CEA-LIST, France.
Fabian Chersi has a BSC and a Master in Electronic Physics, and a PhD in Artificial Intelligence and Robotics. He did a Postdoc at the Institute for Neuroinformatics in Zurich and at the University College of in London on brain inspired AI. He then moved to Paris where he worked on the development of neuromorphic algorithms and ASICs for deep neural networks. He currently works at the Commission for Atomic and Alternative Energies Commission (CEA) in Paris as a senior AI architect on DNN optimization and low power AI chips for the Edge.
17:15-18:00
Course C01-2: Deploying DNNs in RISC-V with Klepsydra AI: A step-by-step, end-to-end workshop. Pablo Ghiglino, CEO, Klepsydra Technologies, Switzerland, Mandar Harshe, Senior Software Developer, Klepsydra Technologies, Switzerland.
Course description:
Klepsydra AI is an inference engine designed to execute pre-trained machine learning models in real time. Developing an end-to-end AI application is complex, involving model design, training, testing, deployment, and validation—often overwhelming for newcomers. To simplify this process, Klepsydra.
Technologies is hosting a workshop focused on Deep Neural Networks (DNNs), the primary interest of the RISC-V and AI communities. Participants will:
• Download a pre-trained DNN for space applications.
• Convert it to ONNX format.
• Deploy and test it on a target board.
• Optimize execution via configuration files.
• Run the optimized inference.
The workshop will share six Raspberry Pi devices with the attendees, and one RISC-V board running Linux will be used by the presenter to follow the same workshop steps, showcasing Klepsydra AI’s versatility and efficiency across different hardware.
Lecturer(s): Pablo Ghiglino, CEO, Klepsydra Technologies, Switzerland, Mandar Harshe, Senior Software Developer, Klepsydra Technologies, Switzerland.
Dr. Pablo Ghiglino: Over 10-year experience in the finance industry where he was exposed to high-speed data-processing environment results in high economic impact. Parallel to his work in finance, Pablo successfully studied a PhD at the Surrey Space Center in Aerospace Engineering in the field of “High Precision Planetary Landing” with five important publications in IEEE and AIAA. After his studies, he got involved in several navigation projects, where he seized the opportunity and created Klepsydra to respond to the needs for cutting Ai for onboard computers by implementing successful software techniques inspired in his work in financial trading.
Dr. Mandar Harshe is a Senior Developer at Klepsydra Technologies GmbH. He obtained his Bachelor of Engineering from the Savitribai Phule Pune University, India and Master of Science (Mechanical Engineering – Dynamics, Systems and Controls) from University of Florida, USA. He further obtained his PhD in Real-time Computing, Robotics and Automation from MINES ParisTech, France working on cable-driven manipulators at INRIA Sophia Antipolis. He has worked at the LORIA lab at Inria Nancy and University of Lorraine, on activity-monitoring for monitoring health indices. He has worked at Wandercraft (Paris) developing medical exoskeletons, working on one of the first self-balanced walking exoskeletons. He also has 2 years of experience working in Deep Learning, working on audio recognition. Since 2019, he works at Klepsydra Technologies AG.
18:00-19:30
Plenary Session D01
18:00-19:30
Session D01 (Room Aula Magna – Scuola Sant’Anna): Edge AI technology trends and roadmap
18:00-18:45
Course D01-1: The intelligent edge – Navigating the frontiers of edge AI technology and applications (Part 1). Ovidiu Vermesan, Chief Scientist, SINTEF, Norway; Giulio Urlini, Research Program Manager, STMicroelectronics, Italy; Alain Pagani, Principal Researcher and Deputy Director, DFKI, Germany.
Course description:
The course explores the rapidly evolving landscape of Edge AI, examining how distributed intelligence is transforming the relationship with technology across physical and virtual domains. Participants will gain insights into cutting-edge research directions and emerging trends that are reshaping the edge AI ecosystem. By bridging technical innovation with practical implementation considerations, the course provides a comprehensive roadmap for researchers, developers, and industry stakeholders navigating the future of edge AI and allow the participants to be involved in the discussions on future trends and developments of edge AI technologies and applications.
Lecturer(s): Ovidiu Vermesan, Chief Scientist, SINTEF, Norway; Giulio Urlini, Research Program Manager, STMicroelectronics, Italy; Alain Pagani, Principal Researcher and Deputy Director, DFKI, Germany.
Ovidiu Vermesan is Chief Scientist at SINTEF Digital, Oslo, where he is involved in applied research on future edge autonomous intelligent systems and edge AI, wireless sensing devices and networks, smart systems integration, microelectronics design of integrated systems (analogue and mixed signal), IIoT. He holds a PhD in Microelectronics and a Master of International Business (MIB).
His applied research activities focus on advancing edge AI processing, embedded electronics, wireless and smart sensing technologies, and the convergence of these technologies and applying the developments to applications such as green mobility, energy, buildings, autonomous systems, electric connected, autonomous, and shared vehicles, and industrial manufacturing. He is currently the coordinator of the Edge AI Technologies for Optimised Performance Embedded Processing (EdgeAI) project.
Giulio Urlini has graduated in electronic engineering in 2001 and worked for STMicroelectronics since 1999. In the past 15 years he has represented ST in many collaborative funded projects in FP7, H2020, Horizon Europe, ECSEL, KDT and Chip JU programs and has been the coordinator of several jointly research initiatives (OPERA, ALOHA, NeuroSoC). He is actively working in the field of Artificial intelligence for Edge devices, and he is coordinating several STMicroelectronics programs for the research in the field of Neuromorphic devices.
Alain Pagani is Principal Researcher and deputy director of the Augmented Vision research department at the German Research Center for Artificial Intelligence (DFKI). His research interests include Artificial Intelligence, Computer Vision, Image Understanding, and Extended Reality. He is the coordinator of the Network of Excellence dAIEDGE about distributed AI at the edge, and the coordinator of the Horizon Europe project CORTEX about remote cooperation using extended reality. He is lecturer at the University of Kaiserslautern-Landau, and he published over 100 articles in conferences and journals. His research finds applications in eXtended Reality for telecooperation (project CORTEX), Artificial Intelligence and Computer Vision for Human-Robot cooperation (project FLUENTLY), Artificial Intelligence and Augmented Reality for analysis of extremely large data (project ExtremeXP).
18:45-19:30
Course D01-2: Panel discussions – Navigating the frontiers of edge AI technology and applications (Part2).
Panel description:
The panel will discuss the technological and applications trends of Edge AI and the road map for future developments in Europe.
Moderator(s): Ovidiu Vermesan, Chief Scientist, SINTEF, Norway; Giulio Urlini, Research Program Manager, STMicroelectronics, Italy; Alain Pagani, Principal Researcher and Deputy Director, DFKI, Germany.
Ovidiu Vermesan is Chief Scientist at SINTEF Digital, Oslo, where he is involved in applied research on future edge autonomous intelligent systems and edge AI, wireless sensing devices and networks, smart systems integration, microelectronics design of integrated systems (analogue and mixed signal), IIoT. He holds a PhD in Microelectronics and a Master of International Business (MIB).
His applied research activities focus on advancing edge AI processing, embedded electronics, wireless and smart sensing technologies, and the convergence of these technologies and applying the developments to applications such as green mobility, energy, buildings, autonomous systems, electric connected, autonomous, and shared vehicles, and industrial manufacturing. He is currently the coordinator of the Edge AI Technologies for Optimised Performance Embedded Processing (EdgeAI) project.
Giulio Urlini has graduated in electronic engineering in 2001 and worked for STMicroelectronics since 1999. In the past 15 years he has represented ST in many collaborative funded projects in FP7, H2020, Horizon Europe, ECSEL, KDT and Chip JU programs and has been the coordinator of several jointly research initiatives (OPERA, ALOHA, NeuroSoC). He is actively working in the field of Artificial intelligence for Edge devices, and he is coordinating several STMicroelectronics programs for the research in the field of Neuromorphic devices.
Alain Pagani is Principal Researcher and deputy director of the Augmented Vision research department at the German Research Center for Artificial Intelligence (DFKI). His research interests include Artificial Intelligence, Computer Vision, Image Understanding, and Extended Reality. He is the coordinator of the Network of Excellence dAIEDGE about distributed AI at the edge, and the coordinator of the Horizon Europe project CORTEX about remote cooperation using extended reality. He is lecturer at the University of Kaiserslautern-Landau, and he published over 100 articles in conferences and journals. His research finds applications in eXtended Reality for telecooperation (project CORTEX), Artificial Intelligence and Computer Vision for Human-Robot cooperation (project FLUENTLY), Artificial Intelligence and Augmented Reality for analysis of extremely large data (project ExtremeXP).
Lecture link:
20:30
Dinner
09:00
Tuesday 8 July 2025 – Day 2
09:00-09:30
Welcome and Registration: TeCIP Institute, Pisa
RISC-V and Edge AI Architectures:
Edge AI concepts and methods:
09:30-11:00
Parallel Sessions A02 & B02
09:30-11:00
Session A02 (Room A – TeCIP Institute): RISC-V
Session B02 (Room B – TeCIP Institute): Cognitive explainable concepts
09:30-10:15
Course A02-1: End-to-end neural networks on RISC-V multi-core processors: from model optimization to full-system simulation. Giuseppe Tagliavini, University of Bologna, Italy; Alessio Burrello, position, Politecnico di Torino, Italy; Daniele Jahier Pagliari, Politecnico di Torino, Italy.
Course description:
In this course, we will cover the optimization of deep neural networks for RISC-V architectures. We will delve into techniques tailored for enhancing neural network execution on low-power RISC-V platforms, examining methods applied before deploying the network onto the hardware, during or post-training. Furthermore, we will describe the automatic compilation of inference code from high-level Python representations of neural networks into optimized binaries specific to RISC-V targets. Lastly, we will introduce an open-source ISS to simulate the execution of neural networks on advanced RISC-V SoCs. The course aims to provide participants with cutting-edge tools and knowledge to optimize and execute end-to-end neural network applications on RISC-V-based devices, enriching their skillset in the field of low-power Edge AI systems design.
Lecturer(s): Giuseppe Tagliavini, University of Bologna, Italy; Alessio Burrello, position, Politecnico di Torino, Italy; Daniele Jahier Pagliari, Politecnico di Torino, Italy.
Giuseppe Tagliavini received a Ph.D. in Electronics, Telecommunications, and Information Technologies Engineering from the University of Bologna (Italy) in 2017. He is currently a Tenure-Track Assistant Professor with the Department of Computer Science and Engineering (DISI) of the University of Bologna. His research interests are mainly focused on programming models, orchestration tools, and compiler optimizations for AI-enabled resource-constrained computing platforms, with a strong emphasis on parallel architectures and accelerators in the context of ultra-low-power IoT end nodes. His expertise in this research field covers multiple levels of the software stack, with the overall objective of optimizing computational effort and energy efficiency. He has published more than 60 papers in peer-reviewed international journals and conferences.
Alessio Burrello is currently a post-doctoral researcher at Politecnico di Torino. He received his M.Sc. and Ph.D. degrees in Electronic Engineering at the Politecnico of Turin, Italy, and the University of Bologna, respectively, in 2018 and 2023. His research interests include parallel programming models for embedded systems, machine and deep learning, hardware-oriented deep learning, and code optimization for multi-core systems. He has published over 70 papers in peer-reviewed international journals and conferences. His work has been awarded different times, including best paper awards at IEEE BioCAS 2018, IEEE ISVLSI 2023, and DATE 2024.
Daniele Jahier Pagliari is currently an Associate Professor of Computer Engineering at Politecnico di Torino, Italy. He received his PhD from the same institution in 2018. He is a member of the IEEE since 2015. His recent research interests are mainly in the optimization of AI applications on embedded and edge devices, with a particular focus on energy-efficiency. He has published over 90 papers in IEEE/ACM conference and journals and his work received a best paper award at the DATE 2024 conference. He has served as track chair for ACM/IEEE ISLPED since 2024 and is an associate editor of the IEEE Internet of Things Journal.
Course B02-1: Cognitive approaches to low-code real-time control of digital twins and swarm monitoring and control. Dave Raggett, W3C/ERCIM, UK.
Course description:
This lecture will introduce cognitive approaches to low-code real-time control for swarms of agents using digital twins as abstractions along with 2.5D and 3D approaches for web-based monitoring and control. Current approaches to industrial robots focus on highly repetitive tasks. Cognitive approaches generalise this to enable dynamic context sensitive switching between multiple behaviours for greater flexibility in human-robot and robot-robot collaborative tasks. Web-based monitoring and control simplifies human stewardship of the factory floor
Lecturer(s): Dave Raggett, W3C/ERCIM, UK.
Dr. Dave Raggett is a web pioneer with a lifelong interest in AI, gaining experience at the University of Oxford (PhD), the Machine Intelligence Research Unit at the University of Edinburgh, the Logic Programming Department at Imperial College, the Computer Science and AI Lab at MIT and many years at HP Lab’s knowledge based programming department. He is now a member of W3C/ERCIM and has been involved in a succession of European projects. He founded W3C’s Cognitive AI Community Group and is driving ongoing work on human-like AI. He holds an honorary professorship at the University of the West of England.
10:15-11:00
Course A02-2: Exploring hardware-efficient architectures for reducing data transfer in DNNs. Mounika Vaddeboina, Senior Engineer System Concept and Architecture, Infineon Technologies AG, Germany; Sebastian Prebeck, Hardware Design Senior Engineer, Infineon Technologies AG, Germany.
Course description:
Domain-specific hardware is emerging as a promising solution amid the slowdown of general-purpose processor improvements. Deep Neural Networks (DNNs) have revolutionized Artificial Intelligence (AI) tasks, but their remarkable accuracy comes at the cost of high memory consumption and computational complexity, which limits deployment in embedded systems. To address these challenges, DNN compression has become a critical approach for memory optimization and computational acceleration. One aspect of the lecture introduces the memory mappings and addressing patterns for inference compute of uncompressed NNs. Another aspect focuses on the workings of a Golomb-Rice (GR) encoding method for NN compression featuring multiple decoder architectures for parallel processing. It extends beyond weight compression to include activation compression, providing a compression-decompression pipeline for DNNs.
Lecturer(s)/Presenter(s): Mounika Vaddeboina, Senior Engineer System Concept and Architecture, Infineon Technologies AG, Germany; Sebastian Prebeck, Hardware Design Senior Engineer, Infineon Technologies AG, Germany.
Mounika Vaddeboina is a Senior Engineer in System Concept and Architecture. Her research focuses on developing innovative solutions for digital design automation and energy-efficient AI inference, with particular emphasis on compression and decompression techniques for Neural Networks.
Dr. -Ing. Sebastian Prebeck is a Hardware Design Senior Engineer. He works on digital hardware design and verification for RISC-V subsystems and RISC-V related IPs. His research focus is on lightweight AI acceleration on small uC based platforms.
Course B02-2: Explainable and generative modelling on the edge for industrial application. Pasquale Coscia, Tenure-Track Researcher, Università degli Studi di Milano, Italy.
Course description:
Industrial AI faces significant challenges, including limited publicly available datasets, sensor-specific variations, and the need to balance model performance with explainability. This course explores state-of-the-art AI techniques, focusing on explainable deep learning and synthetic defect generation for anomaly detection. Given the scarcity of industrial images, it will examine methods for improving feature disentanglement and efficient generative approaches for resource-constrained devices. By integrating these approaches, the course aims to provide effective tools for deploying transparent and efficient AI solutions in industrial settings.
Lecturer(s): Pasquale Coscia, Tenure-Track Researcher, Università degli Studi di Milano, Italy. .
Pasquale Coscia is a Tenure-Track Researcher in the Department of Computer Science at the Università degli Studi di Milano and a member of the Industrial, Environmental and Biometric Informatics (IEBI) Laboratory. He earned his Ph.D. degree (2019) in Industrial and Information Engineering from Università degli Studi della Campania “Luigi Vanvitelli”. From 2019 to 2022, he was a post-doctoral researcher at the Università degli Studi di Padova. In 2024, he was a Visiting Researcher at the University of Toronto, ON, Canada. He is the Co-chair of the Intelligent Measurement Systems Technical Committee (TC-22) of the IEEE Instrumentation and Measurement Society (since 2023). His research activities focus on theoretical, methodological, and applied aspects of computational intelligence for signal and image processing.
11:00-11:30
Morning Coffee/Tea Break
11:30-13:00
Parallel Sessions C02 & D02
11:30-13:00
Session C02 (Room A – TeCIP Institute): Vision models at the edge
Session D02 (Room B – TeCIP Institute): Federated learning
11:30-12:15
Course C02-1: Optimizing vision transformers for the edge: The power of pruning. Martyna Poreba, CEA-List, France; Karim Ben Chehida, CEA-List, France.
Course description:
This talk focuses on optimizing Vision Transformer (ViT) models for resource-constrained edge devices. Existing pruning methods, particularly dynamic pruning, will be classified and explained. This compression technique typically involves adding an intermediate block within a ViT layer that computes metrics (e.g., similarity, informativeness, objectivity) on tokens, deciding at runtime which tokens to eliminate or merge. While the effectiveness of pruning is generally validated in terms of theoretical FLOPs and accuracy, hardware specifications are often overlooked. A list of open research questions will be provided, including pruning performance on edge devices, how pruning can be tailored to the compute and memory limitations, and its integration with other model compression techniques like quantization.
Lecturer(s): Martyna Poreba, CEA-List, France; Karim Ben Chehida, CEA-List, France.
Martyna Poreba is a Research engineer at CEA-List, specializing in multimodal perception for embedded systems. She holds a PhD in Computer Science from MINES Paris, where her early work focused on mobile mapping, photogrammetry, and 3D data processing. She then conducted research on cross-domain and multi- source image retrieval at the French Ministry of Ecology. Currently, her research focuses on optimizing machine learning models for resource-constrained devices, aiming to advance AI applications in real-world scenarios, including image segmentation, object detection and tracking.
Karim Ben Chehida received his PhD degree in Automation, Signal and Image Processing from the University of Nice Sophia Antipolis, France, in 2004. In 2005, he joined the IC Design and Embedded Software Division, CEA LIST (French Atomic Energy Commission), Saclay, France, as a Research Engineer to work on system-level resource and power management for embedded architectures. His current research interests include the compression of perception models and their dynamic inference to adapt to the scene content.
Course D02-1: Privacy in federated learning. Chuan Xu, Assistant professor, Inria/Université Côte d’Azur, France.
Course description:
This course explores privacy attacks and defences in federated learning. We will examine four threat models: when the server or external world is honest-but-curious and when some servers or clients are malicious. Key attack techniques, such as inversion and membership inference attacks, will be presented, along with demonstrations of privacy leakage. The second part of the course focuses on privacy-preserving methods, including differential privacy and secure aggregation. We will discuss their applications in federated learning, analysing how they mitigate privacy risks under different threat models and the trade-offs involved.
Lecturer(s): Chuan Xu, Assistant professor, Inria/Université Côte d’Azur, France.
Chuan Xu is an Assistant Professor at Université Côte
d’Azur, France since September 2021. She earned her Ph.D. from Université Paris-Saclay in 2017, then spent a year as an ATER there. Before joining the Coati Inria project-team, she was a postdoctoral researcher in the NEO Inria project-team. Her research focuses on distributed (federated) learning, distributed algorithms, and data privacy. She is particularly interested in designing efficient attacks and defences for distributed learning systems and exploring alternative algorithms to enhance privacy and robustness guarantees.
12:15-13:00
Course C02-2: Deploying, benchmarking and profiling of large vision models at the edge. Carmelo Scribano, post-doctoral researcher, University of Modena and Reggio Emilia, Italy.
Course description:
This lecture provides an in-depth deep technical exploration of deploying large-scale vision models on edge devices, focusing on NVIDIA hardware such as the NVIDIA Orin family. We will delve into the intricacies of the TensorRT framework, covering advanced optimization techniques such as layer fusion, kernel auto-tuning, and efficient memory management strategies. The session emphasizes rigorous benchmarking, evaluating latency, throughput, and power consumption, and advanced profiling methods to validate execution flow at both the kernel and operator levels. Participants will gain a comprehensive understanding of performance tuning and systematic evaluation, developing technical expertise necessary to implement state-of-the-art vision models on edge platforms.
Lecturer(s): Carmelo Scribano, post-doctoral researcher, University of Modena and Reggio Emilia, Italy.
Dr. Carmelo Scribano is currently a post-doctoral researcher at the University of Modena and Reggio Emilia (Italy). He completed his PhD in mathematics at the University of Parma, in consortium with the University of Modena and the University of Ferrara, under the supervision of Prof. Marko Bertogna. Since 2019, he has been a member of the High- Performance Real-Time (HiPeRT) Lab of University of Modena. Over the years, he has contributed to numerous European projects and industrial collaborations. Dr. Scribano’s research interests encompass deep learning and computer vision techniques tailored for inference on embedded systems. His works focus on reducing the computational footprint of deep learning models and developing computer vision techniques adaptable to the automotive sector. In September 2024 he spent 6 months as visiting post-doctoral researcher at the Institute for Computer Science, Artificial Intelligence and Technology (INSAIT), working with Dr. Danda Pani Paudel and Prof. Luc Van Gool. His research at INSAIT focuses on artificial intelligence for low-power embedded devices as part of the dAIEdge project.
Course D02-2: Federated learning in edge computing. Sevasti Politi, AI/ML Technical Leader, ITML, Greece; Konstantinos Lieros, ML engineer GNT, Greece.
Course description:
The course will cover (i) how FL enables edge intelligence; (ii) how FL implemented in EdgeAI project use cases; (iii) challenges of training on edge devices (resource constraints etc). During this training, we will explore how Federated Learning (FL) empowers edge computing by enabling model training while preserving data privacy. We will discuss its role in EdgeAI applications, key benefits, and real-world use cases. Additionally, we will address the challenges of training on resource-constrained edge devices and potential solutions to overcome these limitations.
Lecturer(s): Sevasti Politi, AI/ML Technical Leader, ITML, Greece; Konstantinos Lieros, ML engineer GNT, Greece.
Sevasti Politi is a machine learning engineer at ITML in Greece, where she also leads the AI/ML Technical Team. She works on multiple EU projects aimed at developing AI-based systems across various domains, such as cybersecurity, automotive, energy sustainability, and mobility as a service. Her responsibilities encompass MLOps, data management pipelines, streaming data pipelines, and model development and optimization. Sevasti holds a Master’s degree in Informatics, focusing on business computing and big data analytics, in addition to an Integrated Master’s degree in Applied Mathematical and Physical Sciences. Her primary interests lie in data management, machine learning, federated learning, edge computing, and code optimization for performance and scalability.
Konstantinos Lieros is a Machine Learning Engineer at GNT in Greece, specializing in designing and implementing scalable systems and advanced machine learning solutions. With a strong focus on building robust microservices, ensuring high performance, reliability, and seamless integration across platforms. Konstantinos has extensive experience in end-to-end machine learning workflows, including data processing, model training, real-time inference, and performance optimization. A graduate in Informatics and Computer Engineering from the University of West Attica, he is skilled in containerization technologies and API development for efficient communication between systems. Additionally, he has worked with distributed systems and cutting-edge methodologies like federated learning, contributing to the development of intelligent and adaptive applications. With a focus on scalability, automation, and system efficiency, he is committed to driving innovation and delivering impactful solutions across diverse projects.
13:00-14:00
Networking Lunch
14:00-15:30
Parallel Sessions E02 & F02
14:00-15:30
Session E02 (Room A – TeCIP Institute): Security and privacy
Session F02 (Room B – TeCIP Institute): Edge AI for smart manufacturing
14:00-14:45
Course E02-1: Privacy enhancing technologies (PETs). Aysajan Abidin, Research Manager, COSIC KU Leuven, Belgium.
Course description:
This course will focus on the roles Privacy Enhancing Technologies (PETs) – such as Multiparty Computation, Fully Homomorphic Encryption, and Zero-Knowledge Proofs – play in achieving the security and privacy of data indistributed AI settings. It starts with a gentle introduction to relevant PETs, discusses some concrete proposals, and sheds light on existing challenges.
Lecturer(s): Aysajan Abidin, Research Manager, COSIC KU Leuven, Belgium.
Aysajan obtained his PhD in information coding with a focus on security of the classical subsystems of Quantum Key Distribution from Linkoping University, Sweden, in 2013. Before joining COSIC at KU Leuven in 2015, he was a postdoctoral researcher at Chalmers University of Technology, Sweden, on privacy-preserving biometrics. His research interests include threshold cryptography,
MPC/FHE applications to privacy-preserving ML/AI, distance bounding, and cryptographic protocols to real-world problems.
Course F02-1: Systematic definition and verification of edge AI requirements. Silke Braun, Senior Manager Innovation & Funding, Infineon Technologies Austria AG, Austria; Michael Cernusca, Project Director for Global Public R&D Funding, ams OSRAM AG, Austria; Roman Lobner, Founder and CEO SoftwareCUBE SCP GmbH, Austria.
Course description:
How can we define and verify meaningful requirements for Edge AI applications on resource-constrained systems? This course equips you with essential skills to specify and validate AI system behaviour through effective requirement definition and verification methodology. Explore common challenges and best practices through real-world examples from the semiconductor industry, including virtual metrology, root cause analysis, and gesture recognition. You’ll gain hands-on experience with a practical, tooling-supported workflow using VeriACT—from defined requirements to verification planning, test execution and result validation. Learn how traceability and real-time dashboards enhance transparency, quality, and control in requirement verification and therefore contributes to reliable Edge AI system development.
Lecturer(s): Silke Braun, Senior Manager Innovation & Funding, Infineon Technologies Austria AG, Austria; Michael Cernusca, Project Director for Global Public R&D Funding, ams OSRAM AG, Austria; Roman Lobner, Founder and CEO SoftwareCUBE SCP GmbH, Austria.
Silke Braun completed her studies in microsystems engineering at Mittweida University of Applied Sciences (Germany). In 2004, she started working at Infineon Technologies Dresden as a defect density and metrology engineer; since 2011 in Villach (Austria). She strengthened her knowledge in the field of quality management. Since 2022, she focuses her work on the field of funding and works in various EU-funded projects.
Dr. Lukas Schafzahl, Metrology Staff Engineer, ams OSRAM AG, Austria
Studied Technical Chemistry at TU Graz and KTH Stockholm, graduated from the Institute of Chemistry and Technology of Materials working on interphase chemistries in Li-ion batteries and beyond. Since 2018, worked at ams OSRAM AG as Application Engineer responsible for spectroscopic measurements of optical filters, critical dimension measurements thin layer characterization.
Roman Lobner: Since founding Company SoftwareCUBE in 2006, Roman Lobner has served as CEO and lead consultant, specializing in high-automation solutions for Industry 4.0. His expertise spans equipment and process automation, system integration, and real-time data acquisition. He has led numerous digitalization projects focused on verification management, ISO compliance and developed mission-critical software systems. With deep industry experience in the semiconductor sector, he manages international projects and software development lifecycles using agile methodologies. His work bridges strategic consulting and hands-on implementation, delivering sustainable innovation across complex, highly regulated environments.
14:45-15:30
Course E02-2: AI under attack: Navigating the security and privacy frontier. Ihsen Alouani, Associate Professor UPHF, France, Senior Lecturer, Queen’s University Belfast.
Course description:
Discover the hidden vulnerabilities in Artificial Intelligence systems and learn to defend against cutting-edge cyber threats! This engaging course provides you with essential insights into AI security and privacy, including adversarial machine learning, model poisoning, privacy-preserving techniques, and trustworthy AI governance. Through interactive sessions, practical exercises, and real-world case studies, you’ll learn to identify weaknesses in AI models, implement robust defences, and protect sensitive data from privacy breaches. Ideal for students interested in trustworthy AI, and data privacy, this course equips you with foundational knowledge to navigate the evolving threat landscape and build secure, privacy-aware AI solutions.
Lecturer(s): Ihsen Alouani, Associate Professor UPHF, France, Senior Lecturer, Queen’s University Belfast.
Ihsen Alouani received his master’s degree from the National Engineering School of Sousse, Tunisia, and completed his Ph.D. at Université Polytechnique Hauts- de-France in 2016. Following his Ph.D., he undertook a postdoctoral position at the Technical University of Munich, Germany. Since 2018, he has been an Associate Professor at the IEMN Laboratory (CNRS) at Université Polytechnique Hauts-de-France. In 2022, he joined Queen’s University Belfast, affiliated with the Centre for Secure Information Technologies (CSIT). Dr. Alouani is the Principal Investigator of several projects from different funding agencies (EPSRC, EU projects, etc). He is serving on the program committees of several international conferences, including the Design Automation Conference (DAC), and Design, Automation and Test in Europe Conference (DATE), etc. His research interests include machine learning security and privacy, approximate computing and Edge AI.
Course F02-2: Powering the future of industry with multimodal large language models: From chip design to smart manufacturing. Jie He, TUWIEN, Austria.
Course description:
Multimodal Large Language Models (MLLMs) are rapidly transforming how we handle complex visual and textual data in engineering. This session explores real-world applications of MLLMs in the semiconductor industry. It will feature a diverse set of case studies, from requirement specification to design and verification of digital circuits, and to process optimization in chip manufacturing. The participants will gain both theoretical insights and hands-on knowledge about MLLMs, such as its general principles and building blocks, the training and evaluation approaches, and strategies to design high- quality industrial datasets. You will also learn how cutting-edge AI techniques like small language models and reinforcement learning are addressing real-world challenges, shaping the future of smarter, faster, and more adaptive engineering workflows.
Lecturer(s): Jie He, TUWIEN, Austria.
Jie He is a PhD student in Informatics from TU Wien. He obtained his master degree in University of Science and Technology of China, and bachelor degree in Shandong University, China. His research interests lie in AI-enabled design automation of embedded systems and cyber-physical systems. He is currently working on automatic interpretation of large-scale timing diagrams and 8D reports with MLLMs, and control of automated driving system under physical limit.
Jie has published several papers in top-tier conferences, such as ICSE’22 (translating English requirements into signal temporal logic), DAC’23 (translating timing diagram pictures into formal specifications), DAC’24 (automatic design of Robotic-IoT systems). He also actively participates in European and Austrian projects such as EdgeAI, iDev40, and ADEX.
15:30-16:00
Afternoon Coffee/Tea Break
16:00-17:30
Parallel Sessions G02 & H02
16:00-17:30
Session G02 (Room A – TeCIP Institute): Edge AI architectures
Session H02 (Room B – TeCIP Institute): Data insights in edge AI
16:00-16:45
Course G02-1: Lowering barriers for neuromorphic computing at the micro-edge. Paolo Meloni, Associate Professor, Dipartimento Ingegneria Elettrica ed Elettronica, Università degli Studi di Cagliari, Italy.
Course description:
This course offers a comprehensive introduction to Spiking Neural Networks (SNNs), emphasizing both theoretical foundations and practical implementations. We begin with the basics of SNNs and explore widely used open-source libraries for training and simulation. The course then surveys the primary classes of hardware suited for SNN deployment, focusing on performance and efficiency trade-offs. A detailed case study presents easy-access implementation strategy. First, we discuss low-end FPGAs, showcasing a cost-effective, scalable approach. Building on this, we examine the potential of open hardware and open PDKs to replicate and extend such designs onto custom chips. Finally, we explore lightweight SNN architectures tailored for biosignal-related applications.
Lecturer(s): Paolo Meloni, Associate Professor, Dipartimento Ingegneria Elettrica ed Elettronica, Università degli Studi di Cagliari, Italy.
Paolo Meloni is currently associate professor at the Department of Electrical and Electronic Engineering (DIEE) at the University of Cagliari. He is author of several international papers and tutor of many bachelor and master students’ thesis in Electronic Engineering. His research activity is mainly focused on the development of advanced digital systems, with emphasis on the application-driven design and programming of SoCs and FPGAs.
Course H02-1: Generative AI to gain data insights from the shopfloor intuitively. Aparna Saisree Thuluva, Siemens, Germany.
Course description:
There exist several machines deployed on the factory shopfloor that produce operational (OT) data. They usually come from diverse vendors, which produce data in diverse data formats, units, semantics etc. It is crucial to normalize the diverse data and provide a harmonized and uniform interface (e.g., Asset Administration Shell (AAS) interface, OPC UA interface etc.) over it so that the OT data can be used consistently and efficiently, and useful data insights can be provided.
In this course we show case how AI technologies such as Large Language Models, Natural Language Processing and Knowledge Graphs can be employed to address these challenges. Firstly, we will show how the above technologies simplify the proces of providing hormonized and uniform interface over OT data. Secondly, we show how complex information from the shopfloor can be easily retrieved and used on the IT layer to gain data insights.
Lecturer(s): Aparna Saisree Thuluva, Siemens, Germany.
Dr. Aparna Thuluva is a senior key expert in the field of IT-OT data integration at Siemens AG, Munich. She focuses on enriching the OT data with semantic and contextual information in order efficiently use the OT data on the IT layer for efficient data insights using AI and Semantic Web technologies.
16:45-17:30
Course G02-2: Computation-in-memory architecture for edge AI applications. Rajendra Bishnoi, Assistant Professor, Delft University of Technology, Netherlands.
Course description:
Neural networks are the foundation of modern artificial intelligence (AI), playing a crucial role in a wide range of cognitive functions, including image classification, voice recognition, self-driving technologies, language understanding, data-driven forecasting, etc. Deploying neural networks on traditional von Neumann architecture hardware, such as CPUs, GPUs, and TPUs, encounters major issues, including poor energy-efficiency and increased latency, primarily due to the memory-wall constraint. Computation-In-Memory (CIM) has the potential to address this problem by integrating processing elements directly into the memory, reducing data movement and enhancing the overall efficiency of the system. The CIM architecture, combined with emerging memristor technologies, delivers various advantages, including leakage-free storage, high density, high scalability, and faster accesses. In this lecture, we will present a comprehensive overview of the memristor-based CIM architecture, including its operational flow and key design principles. We will provide an in-depth analysis of how memristors are utilized to enhance computation efficiency. Additionally, we will explore the intricacies of the hardware- software co-design approach specifically optimized for neural network applications.
Lecturer(s): Rajendra Bishnoi, Assistant Professor, Delft University of Technology, Netherlands.
Rajendra Bishnoi is currently an Assistant Professor with the Computer Engineering Group, Faculty of EEMCS, Delft University of Technology (TU-Delft). He received the Ph.D. degree in computer science from the Karlsruhe Institute of Technology (KIT), Karlsruhe, Germany, in 2017. He was a Research Leader with the MRAM Group, Chair of Dependable Nano Computing, KIT, for more than two years. From 2006 to 2012, he was a Design Engineer with Freescale (now NXP), where he was a part of the Technical Solution Group in memory and SoC flow. His current research interests include neuromorphic computing, computation-in-memory, and emerging technologies. He was a recipient of the EDAA Outstanding Dissertation Award in 2017. He has served on the TPC for several prominent conferences, including DAC’25, DATE’25, DATE’24, ICCAD’24, ETS’24, DATE’23, and DATE’22. In addition, he has held the role of Topic Track Chair for DATE’25, DATE’24, ETS’24, and DSD’23. He is also an editorial board member for the Elsevier Journal and was a guest editor for the Frontiers in Neuroscience Journal in 2022.
Lecture link:
Course H02-2: The role of IoT in edge AI. Jānis Judvaitis, Senior Researcher, Institute of Electronics and Computer Science, Latvia.
Course description:
This course provides an introduction on how the Internet of Things serves as the foundation for Edge AI. The course will explore IoT architectures, sensor networks, data communication protocols and real-time data processing that enable intelligent decision-making at the edge. Overview of lessons learned will be provided from different use cases. Additionally, topics such as IoT security, scalability, and future trends in Edge AI-powered IoT ecosystems will be discussed. Keywords: IoT; edge AI; real time data processing.
Lecturer(s): Jānis Judvaitis, Senior Researcher, Institute of Electronics and Computer Science, Latvia.
Jānis is a senior researcher in Institute of Electronics and Computer Science (EDI) where he has worked from 2013. He is involved in various international projects related to Wireless Sensor Networks, embedded systems, Internet of Things, and edge computing. Expertise in ICT system design, development, implementation, and testing. Author of 20+ scientific articles. Janis defended his doctoral thesis “Enabling Wireless Sensor Network testbeds for high TRL research” in 2023 at the University of Latvia. Research interests include Cyber-Physical Systems, Wireless Sensor Networks, Internet of Things, edge computing and embedded systems.
17:30
End
11:30
Monday 7 July 2025 – Day 1
11:30-12:00
Welcome and Registration: Aula Magna – Scuola Sant’Anna, Pisa
12:00-13:00
Networking Lunch
Edge AI and RISC-V Technology Advancements:
13:00-14:30
Plenary Session A01
13:00-14:30
Session A01 (Room Aula Magna – Scuola Sant’Anna): Edge and generative AI
13:00-14:30
Course A01-1: The path from edge fixed AI to GenAI: Tutorial and practical examples (Part 1 and 2). Danilo Pau, Technical Director, STMicroelectronics, Italy.
Course description:
Artificial Intelligence underwent to several key milestones through the years starting from the vision of Alan Turing, the perceptron of Rosenblatt, the existence theorem of Cybenko and Hornik and the seminal work of G. Hinton on Science. Next AlexNext by winning ILVRC on 2012 started the Fixed Function AI era during which more and more accurate fixed purpose AI workloads were developed and accelerated through GPUs. On 2019 TinyML community was established and the industry started to invest on new edge product culminating on frameworks and mass production low power micro controllers and neural processing units. This plethora of hardware, targeting various industry domains, including a variety of deep learning frameworks, neural network topologies, layers, development environment were the source of the HETEROGENOUS challenge which was a big obstacle to many stakeholders for the mass adoption of AI at the Edge. ST solved that challenge devising and delivering for free the Edge Unified AI Core Technology.
This technology interfaces the most widely used Deep Learning representation standards such as Google Keras, QKeras and Tensorflow Lite and the Open Neural Network Exchange (ONNX). It outputs optimized C code across heterogeneous instruction sets with public APIs for STM32, STM32N6, Stellar MCUs, and MEMs sensors. With that free of charge solution, ST helps the machine learning and embedded developer community to use the IDE they are comfortable with to finalize their application of choice with unprecedented execution speed. This tutorial includes practical demonstration of the tool in action on selected heterogenous hardware will be provided to the audience.
Lecturer(s): Danilo Pau, Technical Director, STMicroelectronics, Italy.
Danilo Pau: As of Dec 2024, I spent 32 years in STMicroelectronics and in System R&D. My (source GoogleScholar) h-index is 30, i10- index 86. I produced 109 invention’s requests, 80 EU and 71 US application patents. I am author of 224 scientific publications, 113 (2012-2015) ISO/IEC/MPEG documents and had 125 invited talks including key notes, seminars, tutorials at Universities/Conferences. I graduated on Electronic Engineering in 1992 at Politecnico di Milano after one year internship in SGS-THOMSON/CASA on 50% memory reduced HD-MAC HW decoder. I have a background on MPEG2 video memory reduction (SD/HD), video (up to H.264) encoding, transcoding, OpenVG/OpenGL-ES since Khronos was established, I co-chaired ISO/IEC SC29-WG11 CDVS and CDVA for 3 years, computer vision and tiny AI since 2016 (through the release of the 2024 Unified AI Core Technology) now part of STM32CubeMX, Stellar-Studio, SPC-Studio.AI, MEMs Studio, STM32 Developer Cloud and the Suite. On 2019 I was elevated IEEE Fellow; on 2022 AAIA Fellow, APSIPA life member since 2020, Sigma Xi honor society (invited) member since 2024. I served as Industry Ambassador (Italy section), coordinator IEEE Region 8 South Europe, vice-chairman of the “Intelligent Cyber-Physical Systems” Task Force IEEE CIS, IEEE R8 AfI internship initiative, AE IEEE TNNLS, Member of the IEEE CTSoc MDAI, 2023 IEEE CS Fellow EVM. I personally curated all the way the IEEE Milestone on Multiple Silicon Technologies on a chip, 1985 ratified by IEEE/2021, and IEEE Milestone on MPEG Multimedia Integrated Circuits, 1984-1993, ratified IEEE/2022, Integrated Circuits for Satellite Digital Radio, 1996-1997, ratified IEEE/2024. I serve TinyML (now EdgeAi) Foundation (Symposium, Summit, EMEA), chair of the TinyML on Device Learning, co-funded AutoTinyML, co-chair TinyML Talks. I am co-chair of EdgeAI Research track at EdgeAI Austin 2025 and co-chair of Edge GenAI working group. Moreover, coordinating contributions to MPAI Neural Watermarking IEEE stardardization. I was awarded with my team “Finmeccanica Innovation Award 2004 Embedded 3D graphics”, STAR Gold award 2014 for high performing Technical Staff (which I co-funded), best demo award with Politecnico di Milano on MPEG Visual Search, on 2014, STAR Silver award on 2019 for STM32Cube.AI, 2020 Spring APSIPA Industrial Distinguished Leaders Award, STAR Innovation Award in 2023. I supervised countless students over 30+ years. I strongly and openly cooperated with many professors, professionals and researchers and love continuing that. Innovation and be part of a scientific community are all my needs.
Lecture link:
14:30-16:00
Plenary Session B01
14:30-16:00
Session B01 (Room Aula Magna – Scuola Sant’Anna): Edge AI and autonomous systems
14:30-15:15
Course B01-1: Scenario generation for autonomous driving using LLMs based on real-world edge data. Andreas Stephan, Robert Bosch, Germany.
Course description:
This course explores scenario generation in the context of autonomous driving. It discusses the role of scenario generation in developing robust, and reliable autonomous driving systems. Participants will learn about common scenario representations and measures how to evaluate them. Furthermore, the course will give a high-level overview of traditional and state-of-the-art scenario generation techniques. These include rule-based methods, statistical methods, deep-learning based methods, and Large Language Model (LLM) driven approaches. The course culminates with an example architecture which leverages edge AI, LLMs and legal specifications to generate critical scenarios according to the provided specification.
Lecturer(s): Andreas Stephan, Robert Bosch, Germany.
Andreas, holds degrees in Mathematics and Computer Science from TU Munich and is a final year PhD candidate in Natural Language Processing (NLP) at the University of Vienna, supervised by Prof. Benjamin Roth. His research focuses on the integration and analysis of noisy sources of information in NLP. Prior to his academic engagement, he was in industry, tackling applied NLP challenges, including information extraction and the integration of graphs with textual data. Currently he works on scenario generation for ADAS systems at Bosch Research. He is broadly interested in NLP’s role in AI and its applications.
Lecture link:
15:15-16:00
Course B01-2: Autonomous mobile robots in smart manufacturing: A real-world approach to edge AI and computer vision. Ayaan Verghese, Research Scientist, Dell Technologies, Ireland; Matthew Keating, Research Scientist, Dell Technologies, Ireland.
Course description:
The course explores the application of AI at the Edge and Photogrammetric Computer Vision for scene understanding to enable autonomous decision making in swarm networks. Currently large conventional factories, have static production lines that may remain unchanged for years. This lecture will demonstrate how centralized fixed manufacturing facilities can evolve using swarms of Autonomous Mobile Robots and smart edge devices communicating and collaborating at the edge, with minimal supervision by humans or central control systems. Students will gain insight into a low-code tool for training computer vision models, design of a reference architecture for edge deployments, the use of intelligent autonomous mobile robots in a manufacturing environment to help implement smart factories.
Lecturer(s): Ayaan Verghese, Research Scientist, Dell Technologies, Ireland; Matthew Keating, Research Scientist, Dell Technologies, Ireland.
Ayaan Verghese is a research scientist at Dell Technologies, exploring the potential of artificial intelligence, computer vision, and robotics to drive innovation and productivity in smart factories. With a strong background in manufacturing processes and industrial automation, he has previously been involved in the development and management of manufacturing processes for enterprise servers and storage systems, as well as deploying automation solutions in production environments. He currently collaborates with academic and industry partners on European Commission Horizon-funded projects, driving knowledge exchange and innovation. Holding a Master’s degree in Electrical and Electronics Engineering, he has published research on Quantum Computing for Electronic Design Automation. Ayaan brings a unique perspective on the intersection of technology and industry.
Matthew Keating is a Research Scientist at Dell Technologies’ Office of the CTO, where he explores the frontiers of Artificial Intelligence, Autonomous Systems, and Semantic Web technologies. With a strong background in industrial research, Matthew has collaborated with academia and industry partners on numerous European Commission Horizon-funded projects, driving innovation and research excellence in Europe. His expertise spans metadata management, machine learning, and earth observation, with a current focus on applying AI to satellite imagery for wildfire tracking and prediction. Matthew is nearing completion of his Master’s thesis on this topic. Matthew brings a unique blend of industry and academic experience, bringing a distinctive perspective to the intersection of technology and societal impact.
Lecture link:
16:00-16:30
Afternoon Coffee/Tea Break
16:30-18:00
Plenary Session C01
16:30-18:00
Session C01 (Room Aula Magna – Scuola Sant’Anna): Edge AI deep neural networks (DNNs) tools
16:30-17:15
Course C01-1: Aidge – a new framework for DNN development, training and deployment on the edge. Fabian Chersi, Senior AI Architect, CEA-LIST, France.
Course description:
The aim of this course is to familiarize the participant with Aidge, a new comprehensive framework developed for simple and efficient DNN importing, construction, optimization, training, testing and deployment. The framework is particularly useful for DNN exploration and optimization, allowing simple and fast prototyping of DNNs with different topologies, quantization levels and optimization methods. It also allows to export to hardware targets such as CPUs, MCUs and GPUs as well as to custom hardware by means of High-Level Synthesis. The framework is developed in C++14 for high performance and provides python bindings for easiness and compatibility with major frameworks such as Pytorch, Tensorflow, ONNX runtime. The course comprises an introductory part and practical part.
Lecturer(s): Fabian Chersi, Senior AI Architect, CEA-LIST, France.
Fabian Chersi has a BSC and a Master in Electronic Physics, and a PhD in Artificial Intelligence and Robotics. He did a Postdoc at the Institute for Neuroinformatics in Zurich and at the University College of in London on brain inspired AI. He then moved to Paris where he worked on the development of neuromorphic algorithms and ASICs for deep neural networks. He currently works at the Commission for Atomic and Alternative Energies Commission (CEA) in Paris as a senior AI architect on DNN optimization and low power AI chips for the Edge.
Lecture link:
17:15-18:00
Course C01-2: Deploying DNNs in RISC-V with Klepsydra AI: A step-by-step, end-to-end workshop. Pablo Ghiglino, CEO, Klepsydra Technologies, Switzerland, Mandar Harshe, Senior Software Developer, Klepsydra Technologies, Switzerland.
Course description:
Klepsydra AI is an inference engine designed to execute pre-trained machine learning models in real time. Developing an end-to-end AI application is complex, involving model design, training, testing, deployment, and validation—often overwhelming for newcomers. To simplify this process, Klepsydra.
Technologies is hosting a workshop focused on Deep Neural Networks (DNNs), the primary interest of the RISC-V and AI communities. Participants will:
• Download a pre-trained DNN for space applications.
• Convert it to ONNX format.
• Deploy and test it on a target board.
• Optimize execution via configuration files.
• Run the optimized inference.
The workshop will share six Raspberry Pi devices with the attendees, and one RISC-V board running Linux will be used by the presenter to follow the same workshop steps, showcasing Klepsydra AI’s versatility and efficiency across different hardware.
Lecturer(s): Pablo Ghiglino, CEO, Klepsydra Technologies, Switzerland, Mandar Harshe, Senior Software Developer, Klepsydra Technologies, Switzerland.
Dr. Pablo Ghiglino: Over 10-year experience in the finance industry where he was exposed to high-speed data-processing environment results in high economic impact. Parallel to his work in finance, Pablo successfully studied a PhD at the Surrey Space Center in Aerospace Engineering in the field of “High Precision Planetary Landing” with five important publications in IEEE and AIAA. After his studies, he got involved in several navigation projects, where he seized the opportunity and created Klepsydra to respond to the needs for cutting Ai for onboard computers by implementing successful software techniques inspired in his work in financial trading.
Dr. Mandar Harshe is a Senior Developer at Klepsydra Technologies GmbH. He obtained his Bachelor of Engineering from the Savitribai Phule Pune University, India and Master of Science (Mechanical Engineering – Dynamics, Systems and Controls) from University of Florida, USA. He further obtained his PhD in Real-time Computing, Robotics and Automation from MINES ParisTech, France working on cable-driven manipulators at INRIA Sophia Antipolis. He has worked at the LORIA lab at Inria Nancy and University of Lorraine, on activity-monitoring for monitoring health indices. He has worked at Wandercraft (Paris) developing medical exoskeletons, working on one of the first self-balanced walking exoskeletons. He also has 2 years of experience working in Deep Learning, working on audio recognition. Since 2019, he works at Klepsydra Technologies AG.
Lecture link:
18:00-19:30
Plenary Session D01
18:00-19:30
Session D01 (Room Aula Magna – Scuola Sant’Anna): Edge AI technology trends and roadmap
18:00-18:45
Course D01-1: The intelligent edge – Navigating the frontiers of edge AI technology and applications (Part 1). Ovidiu Vermesan, Chief Scientist, SINTEF, Norway; Giulio Urlini, Research Program Manager, STMicroelectronics, Italy; Alain Pagani, Principal Researcher and Deputy Director, DFKI, Germany.
Course description:
The course explores the rapidly evolving landscape of Edge AI, examining how distributed intelligence is transforming the relationship with technology across physical and virtual domains. Participants will gain insights into cutting-edge research directions and emerging trends that are reshaping the edge AI ecosystem. By bridging technical innovation with practical implementation considerations, the course provides a comprehensive roadmap for researchers, developers, and industry stakeholders navigating the future of edge AI and allow the participants to be involved in the discussions on future trends and developments of edge AI technologies and applications.
Lecturer(s): Ovidiu Vermesan, Chief Scientist, SINTEF, Norway; Giulio Urlini, Research Program Manager, STMicroelectronics, Italy; Alain Pagani, Principal Researcher and Deputy Director, DFKI, Germany.
Ovidiu Vermesan is Chief Scientist at SINTEF Digital, Oslo, where he is involved in applied research on future edge autonomous intelligent systems and edge AI, wireless sensing devices and networks, smart systems integration, microelectronics design of integrated systems (analogue and mixed signal), IIoT. He holds a PhD in Microelectronics and a Master of International Business (MIB).
His applied research activities focus on advancing edge AI processing, embedded electronics, wireless and smart sensing technologies, and the convergence of these technologies and applying the developments to applications such as green mobility, energy, buildings, autonomous systems, electric connected, autonomous, and shared vehicles, and industrial manufacturing. He is currently the coordinator of the Edge AI Technologies for Optimised Performance Embedded Processing (EdgeAI) project.
Giulio Urlini has graduated in electronic engineering in 2001 and worked for STMicroelectronics since 1999. In the past 15 years he has represented ST in many collaborative funded projects in FP7, H2020, Horizon Europe, ECSEL, KDT and Chip JU programs and has been the coordinator of several jointly research initiatives (OPERA, ALOHA, NeuroSoC). He is actively working in the field of Artificial intelligence for Edge devices, and he is coordinating several STMicroelectronics programs for the research in the field of Neuromorphic devices.
Alain Pagani is Principal Researcher and deputy director of the Augmented Vision research department at the German Research Center for Artificial Intelligence (DFKI). His research interests include Artificial Intelligence, Computer Vision, Image Understanding, and Extended Reality. He is the coordinator of the Network of Excellence dAIEDGE about distributed AI at the edge, and the coordinator of the Horizon Europe project CORTEX about remote cooperation using extended reality. He is lecturer at the University of Kaiserslautern-Landau, and he published over 100 articles in conferences and journals. His research finds applications in eXtended Reality for telecooperation (project CORTEX), Artificial Intelligence and Computer Vision for Human-Robot cooperation (project FLUENTLY), Artificial Intelligence and Augmented Reality for analysis of extremely large data (project ExtremeXP).
Lecture link:
18:45-19:30
Course D01-2: Panel discussions – Navigating the frontiers of edge AI technology and applications (Part2).
Panel description:
The panel will discuss the technological and applications trends of Edge AI and the road map for future developments in Europe.
Moderator(s): Ovidiu Vermesan, Chief Scientist, SINTEF, Norway; Giulio Urlini, Research Program Manager, STMicroelectronics, Italy; Alain Pagani, Principal Researcher and Deputy Director, DFKI, Germany.
Ovidiu Vermesan is Chief Scientist at SINTEF Digital, Oslo, where he is involved in applied research on future edge autonomous intelligent systems and edge AI, wireless sensing devices and networks, smart systems integration, microelectronics design of integrated systems (analogue and mixed signal), IIoT. He holds a PhD in Microelectronics and a Master of International Business (MIB).
His applied research activities focus on advancing edge AI processing, embedded electronics, wireless and smart sensing technologies, and the convergence of these technologies and applying the developments to applications such as green mobility, energy, buildings, autonomous systems, electric connected, autonomous, and shared vehicles, and industrial manufacturing. He is currently the coordinator of the Edge AI Technologies for Optimised Performance Embedded Processing (EdgeAI) project.
Giulio Urlini has graduated in electronic engineering in 2001 and worked for STMicroelectronics since 1999. In the past 15 years he has represented ST in many collaborative funded projects in FP7, H2020, Horizon Europe, ECSEL, KDT and Chip JU programs and has been the coordinator of several jointly research initiatives (OPERA, ALOHA, NeuroSoC). He is actively working in the field of Artificial intelligence for Edge devices, and he is coordinating several STMicroelectronics programs for the research in the field of Neuromorphic devices.
Alain Pagani is Principal Researcher and deputy director of the Augmented Vision research department at the German Research Center for Artificial Intelligence (DFKI). His research interests include Artificial Intelligence, Computer Vision, Image Understanding, and Extended Reality. He is the coordinator of the Network of Excellence dAIEDGE about distributed AI at the edge, and the coordinator of the Horizon Europe project CORTEX about remote cooperation using extended reality. He is lecturer at the University of Kaiserslautern-Landau, and he published over 100 articles in conferences and journals. His research finds applications in eXtended Reality for telecooperation (project CORTEX), Artificial Intelligence and Computer Vision for Human-Robot cooperation (project FLUENTLY), Artificial Intelligence and Augmented Reality for analysis of extremely large data (project ExtremeXP).
Lecture link:
20:30
Dinner
09:00
Tuesday 8 July 2025 – Day 2
09:00-09:30
Welcome and Registration: TeCIP Institute, Pisa
RISC-V and Edge AI Architectures:
Edge AI concepts and methods:
09:30-11:00
Parallel Sessions A02 & B02
09:30-11:00
Session A02 (Room A – TeCIP Institute): RISC-V
Session B02 (Room B – TeCIP Institute): Cognitive explainable concepts
09:30-10:15
Course A02-1: End-to-end neural networks on RISC-V multi-core processors: from model optimization to full-system simulation. Giuseppe Tagliavini, University of Bologna, Italy; Alessio Burrello, position, Politecnico di Torino, Italy; Daniele Jahier Pagliari, Politecnico di Torino, Italy.
Course description:
In this course, we will cover the optimization of deep neural networks for RISC-V architectures. We will delve into techniques tailored for enhancing neural network execution on low-power RISC-V platforms, examining methods applied before deploying the network onto the hardware, during or post-training. Furthermore, we will describe the automatic compilation of inference code from high-level Python representations of neural networks into optimized binaries specific to RISC-V targets. Lastly, we will introduce an open-source ISS to simulate the execution of neural networks on advanced RISC-V SoCs. The course aims to provide participants with cutting-edge tools and knowledge to optimize and execute end-to-end neural network applications on RISC-V-based devices, enriching their skillset in the field of low-power Edge AI systems design.
Lecturer(s): Giuseppe Tagliavini, University of Bologna, Italy; Alessio Burrello, position, Politecnico di Torino, Italy; Daniele Jahier Pagliari, Politecnico di Torino, Italy.
Giuseppe Tagliavini received a Ph.D. in Electronics, Telecommunications, and Information Technologies Engineering from the University of Bologna (Italy) in 2017. He is currently a Tenure-Track Assistant Professor with the Department of Computer Science and Engineering (DISI) of the University of Bologna. His research interests are mainly focused on programming models, orchestration tools, and compiler optimizations for AI-enabled resource-constrained computing platforms, with a strong emphasis on parallel architectures and accelerators in the context of ultra-low-power IoT end nodes. His expertise in this research field covers multiple levels of the software stack, with the overall objective of optimizing computational effort and energy efficiency. He has published more than 60 papers in peer-reviewed international journals and conferences.
Alessio Burrello is currently a post-doctoral researcher at Politecnico di Torino. He received his M.Sc. and Ph.D. degrees in Electronic Engineering at the Politecnico of Turin, Italy, and the University of Bologna, respectively, in 2018 and 2023. His research interests include parallel programming models for embedded systems, machine and deep learning, hardware-oriented deep learning, and code optimization for multi-core systems. He has published over 70 papers in peer-reviewed international journals and conferences. His work has been awarded different times, including best paper awards at IEEE BioCAS 2018, IEEE ISVLSI 2023, and DATE 2024.
Daniele Jahier Pagliari is currently an Associate Professor of Computer Engineering at Politecnico di Torino, Italy. He received his PhD from the same institution in 2018. He is a member of the IEEE since 2015. His recent research interests are mainly in the optimization of AI applications on embedded and edge devices, with a particular focus on energy-efficiency. He has published over 90 papers in IEEE/ACM conference and journals and his work received a best paper award at the DATE 2024 conference. He has served as track chair for ACM/IEEE ISLPED since 2024 and is an associate editor of the IEEE Internet of Things Journal.
Lecture link:
Course B02-1: Cognitive approaches to low-code real-time control of digital twins and swarm monitoring and control. Dave Raggett, W3C/ERCIM, UK.
Course description:
This lecture will introduce cognitive approaches to low-code real-time control for swarms of agents using digital twins as abstractions along with 2.5D and 3D approaches for web-based monitoring and control. Current approaches to industrial robots focus on highly repetitive tasks. Cognitive approaches generalise this to enable dynamic context sensitive switching between multiple behaviours for greater flexibility in human-robot and robot-robot collaborative tasks. Web-based monitoring and control simplifies human stewardship of the factory floor
Lecturer(s): Dave Raggett, W3C/ERCIM, UK.
Dr. Dave Raggett is a web pioneer with a lifelong interest in AI, gaining experience at the University of Oxford (PhD), the Machine Intelligence Research Unit at the University of Edinburgh, the Logic Programming Department at Imperial College, the Computer Science and AI Lab at MIT and many years at HP Lab’s knowledge based programming department. He is now a member of W3C/ERCIM and has been involved in a succession of European projects. He founded W3C’s Cognitive AI Community Group and is driving ongoing work on human-like AI. He holds an honorary professorship at the University of the West of England.
Lecture link
10:15-11:00
Course A02-2: Exploring hardware-efficient architectures for reducing data transfer in DNNs. Mounika Vaddeboina, Senior Engineer System Concept and Architecture, Infineon Technologies AG, Germany; Sebastian Prebeck, Hardware Design Senior Engineer, Infineon Technologies AG, Germany.
Course description:
Domain-specific hardware is emerging as a promising solution amid the slowdown of general-purpose processor improvements. Deep Neural Networks (DNNs) have revolutionized Artificial Intelligence (AI) tasks, but their remarkable accuracy comes at the cost of high memory consumption and computational complexity, which limits deployment in embedded systems. To address these challenges, DNN compression has become a critical approach for memory optimization and computational acceleration. One aspect of the lecture introduces the memory mappings and addressing patterns for inference compute of uncompressed NNs. Another aspect focuses on the workings of a Golomb-Rice (GR) encoding method for NN compression featuring multiple decoder architectures for parallel processing. It extends beyond weight compression to include activation compression, providing a compression-decompression pipeline for DNNs.
Lecturer(s)/Presenter(s): Mounika Vaddeboina, Senior Engineer System Concept and Architecture, Infineon Technologies AG, Germany; Sebastian Prebeck, Hardware Design Senior Engineer, Infineon Technologies AG, Germany.
Mounika Vaddeboina is a Senior Engineer in System Concept and Architecture. Her research focuses on developing innovative solutions for digital design automation and energy-efficient AI inference, with particular emphasis on compression and decompression techniques for Neural Networks.
Dr. -Ing. Sebastian Prebeck is a Hardware Design Senior Engineer. He works on digital hardware design and verification for RISC-V subsystems and RISC-V related IPs. His research focus is on lightweight AI acceleration on small uC based platforms.
Lecture link:
Course B02-2: Explainable and generative modelling on the edge for industrial application. Pasquale Coscia, Tenure-Track Researcher, Università degli Studi di Milano, Italy.
Course description:
Industrial AI faces significant challenges, including limited publicly available datasets, sensor-specific variations, and the need to balance model performance with explainability. This course explores state-of-the-art AI techniques, focusing on explainable deep learning and synthetic defect generation for anomaly detection. Given the scarcity of industrial images, it will examine methods for improving feature disentanglement and efficient generative approaches for resource-constrained devices. By integrating these approaches, the course aims to provide effective tools for deploying transparent and efficient AI solutions in industrial settings.
Lecturer(s): Pasquale Coscia, Tenure-Track Researcher, Università degli Studi di Milano, Italy. .
Pasquale Coscia is a Tenure-Track Researcher in the Department of Computer Science at the Università degli Studi di Milano and a member of the Industrial, Environmental and Biometric Informatics (IEBI) Laboratory. He earned his Ph.D. degree (2019) in Industrial and Information Engineering from Università degli Studi della Campania “Luigi Vanvitelli”. From 2019 to 2022, he was a post-doctoral researcher at the Università degli Studi di Padova. In 2024, he was a Visiting Researcher at the University of Toronto, ON, Canada. He is the Co-chair of the Intelligent Measurement Systems Technical Committee (TC-22) of the IEEE Instrumentation and Measurement Society (since 2023). His research activities focus on theoretical, methodological, and applied aspects of computational intelligence for signal and image processing.
Lecture link
11:00-11:30
Morning Coffee/Tea Break
11:30-13:00
Parallel Sessions C02 & D02
11:30-13:00
Session C02 (Room A – TeCIP Institute): Vision models at the edge
Session D02 (Room B – TeCIP Institute): Federated learning
11:30-12:15
Course C02-1: Optimizing vision transformers for the edge: The power of pruning. Martyna Poreba, CEA-List, France; Karim Ben Chehida, CEA-List, France.
Course description:
This talk focuses on optimizing Vision Transformer (ViT) models for resource-constrained edge devices. Existing pruning methods, particularly dynamic pruning, will be classified and explained. This compression technique typically involves adding an intermediate block within a ViT layer that computes metrics (e.g., similarity, informativeness, objectivity) on tokens, deciding at runtime which tokens to eliminate or merge. While the effectiveness of pruning is generally validated in terms of theoretical FLOPs and accuracy, hardware specifications are often overlooked. A list of open research questions will be provided, including pruning performance on edge devices, how pruning can be tailored to the compute and memory limitations, and its integration with other model compression techniques like quantization.
Lecturer(s): Martyna Poreba, CEA-List, France; Karim Ben Chehida, CEA-List, France.
Martyna Poreba is a Research engineer at CEA-List, specializing in multimodal perception for embedded systems. She holds a PhD in Computer Science from MINES Paris, where her early work focused on mobile mapping, photogrammetry, and 3D data processing. She then conducted research on cross-domain and multi- source image retrieval at the French Ministry of Ecology. Currently, her research focuses on optimizing machine learning models for resource-constrained devices, aiming to advance AI applications in real-world scenarios, including image segmentation, object detection and tracking.
Karim Ben Chehida received his PhD degree in Automation, Signal and Image Processing from the University of Nice Sophia Antipolis, France, in 2004. In 2005, he joined the IC Design and Embedded Software Division, CEA LIST (French Atomic Energy Commission), Saclay, France, as a Research Engineer to work on system-level resource and power management for embedded architectures. His current research interests include the compression of perception models and their dynamic inference to adapt to the scene content.
Lecture link:
Course D02-1: Privacy in federated learning. Chuan Xu, Assistant professor, Inria/Université Côte d’Azur, France.
Course description:
This course explores privacy attacks and defences in federated learning. We will examine four threat models: when the server or external world is honest-but-curious and when some servers or clients are malicious. Key attack techniques, such as inversion and membership inference attacks, will be presented, along with demonstrations of privacy leakage. The second part of the course focuses on privacy-preserving methods, including differential privacy and secure aggregation. We will discuss their applications in federated learning, analysing how they mitigate privacy risks under different threat models and the trade-offs involved.
Lecturer(s): Chuan Xu, Assistant professor, Inria/Université Côte d’Azur, France.
Chuan Xu is an Assistant Professor at Université Côte
d’Azur, France since September 2021. She earned her Ph.D. from Université Paris-Saclay in 2017, then spent a year as an ATER there. Before joining the Coati Inria project-team, she was a postdoctoral researcher in the NEO Inria project-team. Her research focuses on distributed (federated) learning, distributed algorithms, and data privacy. She is particularly interested in designing efficient attacks and defences for distributed learning systems and exploring alternative algorithms to enhance privacy and robustness guarantees.
Lecture link
12:15-13:00
Course C02-2: Deploying, benchmarking and profiling of large vision models at the edge. Carmelo Scribano, post-doctoral researcher, University of Modena and Reggio Emilia, Italy.
Course description:
This lecture provides an in-depth deep technical exploration of deploying large-scale vision models on edge devices, focusing on NVIDIA hardware such as the NVIDIA Orin family. We will delve into the intricacies of the TensorRT framework, covering advanced optimization techniques such as layer fusion, kernel auto-tuning, and efficient memory management strategies. The session emphasizes rigorous benchmarking, evaluating latency, throughput, and power consumption, and advanced profiling methods to validate execution flow at both the kernel and operator levels. Participants will gain a comprehensive understanding of performance tuning and systematic evaluation, developing technical expertise necessary to implement state-of-the-art vision models on edge platforms.
Lecturer(s): Carmelo Scribano, post-doctoral researcher, University of Modena and Reggio Emilia, Italy.
Dr. Carmelo Scribano is currently a post-doctoral researcher at the University of Modena and Reggio Emilia (Italy). He completed his PhD in mathematics at the University of Parma, in consortium with the University of Modena and the University of Ferrara, under the supervision of Prof. Marko Bertogna. Since 2019, he has been a member of the High- Performance Real-Time (HiPeRT) Lab of University of Modena. Over the years, he has contributed to numerous European projects and industrial collaborations. Dr. Scribano’s research interests encompass deep learning and computer vision techniques tailored for inference on embedded systems. His works focus on reducing the computational footprint of deep learning models and developing computer vision techniques adaptable to the automotive sector. In September 2024 he spent 6 months as visiting post-doctoral researcher at the Institute for Computer Science, Artificial Intelligence and Technology (INSAIT), working with Dr. Danda Pani Paudel and Prof. Luc Van Gool. His research at INSAIT focuses on artificial intelligence for low-power embedded devices as part of the dAIEdge project.
Lecture link:
Course D02-2: Federated learning in edge computing. Sevasti Politi, AI/ML Technical Leader, ITML, Greece; Konstantinos Lieros, ML engineer GNT, Greece.
Course description:
The course will cover (i) how FL enables edge intelligence; (ii) how FL implemented in EdgeAI project use cases; (iii) challenges of training on edge devices (resource constraints etc). During this training, we will explore how Federated Learning (FL) empowers edge computing by enabling model training while preserving data privacy. We will discuss its role in EdgeAI applications, key benefits, and real-world use cases. Additionally, we will address the challenges of training on resource-constrained edge devices and potential solutions to overcome these limitations.
Lecturer(s): Sevasti Politi, AI/ML Technical Leader, ITML, Greece; Konstantinos Lieros, ML engineer GNT, Greece.
Sevasti Politi is a machine learning engineer at ITML in Greece, where she also leads the AI/ML Technical Team. She works on multiple EU projects aimed at developing AI-based systems across various domains, such as cybersecurity, automotive, energy sustainability, and mobility as a service. Her responsibilities encompass MLOps, data management pipelines, streaming data pipelines, and model development and optimization. Sevasti holds a Master’s degree in Informatics, focusing on business computing and big data analytics, in addition to an Integrated Master’s degree in Applied Mathematical and Physical Sciences. Her primary interests lie in data management, machine learning, federated learning, edge computing, and code optimization for performance and scalability.
Konstantinos Lieros is a Machine Learning Engineer at GNT in Greece, specializing in designing and implementing scalable systems and advanced machine learning solutions. With a strong focus on building robust microservices, ensuring high performance, reliability, and seamless integration across platforms. Konstantinos has extensive experience in end-to-end machine learning workflows, including data processing, model training, real-time inference, and performance optimization. A graduate in Informatics and Computer Engineering from the University of West Attica, he is skilled in containerization technologies and API development for efficient communication between systems. Additionally, he has worked with distributed systems and cutting-edge methodologies like federated learning, contributing to the development of intelligent and adaptive applications. With a focus on scalability, automation, and system efficiency, he is committed to driving innovation and delivering impactful solutions across diverse projects.
Lecture link
13:00-14:00
Networking Lunch
14:00-15:30
Parallel Sessions E02 & F02
14:00-15:30
Session E02 (Room A – TeCIP Institute): Security and privacy
Session F02 (Room B – TeCIP Institute): Edge AI for smart manufacturing
14:00-14:45
Course E02-1: Privacy enhancing technologies (PETs). Aysajan Abidin, Research Manager, COSIC KU Leuven, Belgium.
Course description:
This course will focus on the roles Privacy Enhancing Technologies (PETs) – such as Multiparty Computation, Fully Homomorphic Encryption, and Zero-Knowledge Proofs – play in achieving the security and privacy of data indistributed AI settings. It starts with a gentle introduction to relevant PETs, discusses some concrete proposals, and sheds light on existing challenges.
Lecturer(s): Aysajan Abidin, Research Manager, COSIC KU Leuven, Belgium.
Aysajan obtained his PhD in information coding with a focus on security of the classical subsystems of Quantum Key Distribution from Linkoping University, Sweden, in 2013. Before joining COSIC at KU Leuven in 2015, he was a postdoctoral researcher at Chalmers University of Technology, Sweden, on privacy-preserving biometrics. His research interests include threshold cryptography,
MPC/FHE applications to privacy-preserving ML/AI, distance bounding, and cryptographic protocols to real-world problems.
Lecture link:
Course F02-1: Systematic definition and verification of edge AI requirements. Silke Braun, Senior Manager Innovation & Funding, Infineon Technologies Austria AG, Austria; Michael Cernusca, Project Director for Global Public R&D Funding, ams OSRAM AG, Austria; Roman Lobner, Founder and CEO SoftwareCUBE SCP GmbH, Austria.
Course description:
How can we define and verify meaningful requirements for Edge AI applications on resource-constrained systems? This course equips you with essential skills to specify and validate AI system behaviour through effective requirement definition and verification methodology. Explore common challenges and best practices through real-world examples from the semiconductor industry, including virtual metrology, root cause analysis, and gesture recognition. You’ll gain hands-on experience with a practical, tooling-supported workflow using VeriACT—from defined requirements to verification planning, test execution and result validation. Learn how traceability and real-time dashboards enhance transparency, quality, and control in requirement verification and therefore contributes to reliable Edge AI system development.
Lecturer(s): Silke Braun, Senior Manager Innovation & Funding, Infineon Technologies Austria AG, Austria; Michael Cernusca, Project Director for Global Public R&D Funding, ams OSRAM AG, Austria; Roman Lobner, Founder and CEO SoftwareCUBE SCP GmbH, Austria.
Silke Braun completed her studies in microsystems engineering at Mittweida University of Applied Sciences (Germany). In 2004, she started working at Infineon Technologies Dresden as a defect density and metrology engineer; since 2011 in Villach (Austria). She strengthened her knowledge in the field of quality management. Since 2022, she focuses her work on the field of funding and works in various EU-funded projects.
Michael Cernusca graduated from the Faculty of Electrical Engineering, TU Wien. From 1998 to 2006, worked at Philips Semiconductors in Product Management, focusing on RFID technology and smart label assembly. Between 2006 and 2018, held roles at AVL List GmbH as Group Product Manager and Global Business Segment Manager, responsible for consumption measurement devices. From 2018 to 2022, served as Director of Product Management and Director of System Innovation & Exploration at ams OSRAM AG, leading teams in product strategy, roadmapping, and innovation. Since 2023, Project Director for Global Public R&D Funding at ams OSRAM AG, managing research collaborations and funding applications.
Roman Lobner: Since founding Company SoftwareCUBE in 2006, Roman Lobner has served as CEO and lead consultant, specializing in high-automation solutions for Industry 4.0. His expertise spans equipment and process automation, system integration, and real-time data acquisition. He has led numerous digitalization projects focused on verification management, ISO compliance and developed mission-critical software systems. With deep industry experience in the semiconductor sector, he manages international projects and software development lifecycles using agile methodologies. His work bridges strategic consulting and hands-on implementation, delivering sustainable innovation across complex, highly regulated environments.
Lecture link
14:45-15:30
Course E02-2: AI under attack: Navigating the security and privacy frontier. Ihsen Alouani, Associate Professor UPHF, France, Senior Lecturer, Queen’s University Belfast.
Course description:
Discover the hidden vulnerabilities in Artificial Intelligence systems and learn to defend against cutting-edge cyber threats! This engaging course provides you with essential insights into AI security and privacy, including adversarial machine learning, model poisoning, privacy-preserving techniques, and trustworthy AI governance. Through interactive sessions, practical exercises, and real-world case studies, you’ll learn to identify weaknesses in AI models, implement robust defences, and protect sensitive data from privacy breaches. Ideal for students interested in trustworthy AI, and data privacy, this course equips you with foundational knowledge to navigate the evolving threat landscape and build secure, privacy-aware AI solutions.
Lecturer(s): Ihsen Alouani, Associate Professor UPHF, France, Senior Lecturer, Queen’s University Belfast.
Ihsen Alouani received his master’s degree from the National Engineering School of Sousse, Tunisia, and completed his Ph.D. at Université Polytechnique Hauts- de-France in 2016. Following his Ph.D., he undertook a postdoctoral position at the Technical University of Munich, Germany. Since 2018, he has been an Associate Professor at the IEMN Laboratory (CNRS) at Université Polytechnique Hauts-de-France. In 2022, he joined Queen’s University Belfast, affiliated with the Centre for Secure Information Technologies (CSIT). Dr. Alouani is the Principal Investigator of several projects from different funding agencies (EPSRC, EU projects, etc). He is serving on the program committees of several international conferences, including the Design Automation Conference (DAC), and Design, Automation and Test in Europe Conference (DATE), etc. His research interests include machine learning security and privacy, approximate computing and Edge AI.
Lecture link:
Course F02-2: Powering the future of industry with multimodal large language models: From chip design to smart manufacturing. Jie He, TUWIEN, Austria.
Course description:
Multimodal Large Language Models (MLLMs) are rapidly transforming how we handle complex visual and textual data in engineering. This session explores real-world applications of MLLMs in the semiconductor industry. It will feature a diverse set of case studies, from requirement specification to design and verification of digital circuits, and to process optimization in chip manufacturing. The participants will gain both theoretical insights and hands-on knowledge about MLLMs, such as its general principles and building blocks, the training and evaluation approaches, and strategies to design high- quality industrial datasets. You will also learn how cutting-edge AI techniques like small language models and reinforcement learning are addressing real-world challenges, shaping the future of smarter, faster, and more adaptive engineering workflows.
Lecturer(s): Jie He, TUWIEN, Austria.
Jie He is a PhD student in Informatics from TU Wien. He obtained his master degree in University of Science and Technology of China, and bachelor degree in Shandong University, China. His research interests lie in AI-enabled design automation of embedded systems and cyber-physical systems. He is currently working on automatic interpretation of large-scale timing diagrams and 8D reports with MLLMs, and control of automated driving system under physical limit.
Jie has published several papers in top-tier conferences, such as ICSE’22 (translating English requirements into signal temporal logic), DAC’23 (translating timing diagram pictures into formal specifications), DAC’24 (automatic design of Robotic-IoT systems). He also actively participates in European and Austrian projects such as EdgeAI, iDev40, and ADEX.
Lecture link
15:30-16:00
Afternoon Coffee/Tea Break
16:00-17:30
Parallel Sessions G02 & H02
16:00-17:30
Session G02 (Room A – TeCIP Institute): Edge AI architectures
Session H02 (Room B – TeCIP Institute): Data insights in edge AI
16:00-16:45
Course G02-1: Lowering barriers for neuromorphic computing at the micro-edge. Paolo Meloni, Associate Professor, Dipartimento Ingegneria Elettrica ed Elettronica, Università degli Studi di Cagliari, Italy.
Course description:
This course offers a comprehensive introduction to Spiking Neural Networks (SNNs), emphasizing both theoretical foundations and practical implementations. We begin with the basics of SNNs and explore widely used open-source libraries for training and simulation. The course then surveys the primary classes of hardware suited for SNN deployment, focusing on performance and efficiency trade-offs. A detailed case study presents easy-access implementation strategy. First, we discuss low-end FPGAs, showcasing a cost-effective, scalable approach. Building on this, we examine the potential of open hardware and open PDKs to replicate and extend such designs onto custom chips. Finally, we explore lightweight SNN architectures tailored for biosignal-related applications.
Lecturer(s): Paolo Meloni, Associate Professor, Dipartimento Ingegneria Elettrica ed Elettronica, Università degli Studi di Cagliari, Italy.
Paolo Meloni is currently associate professor at the Department of Electrical and Electronic Engineering (DIEE) at the University of Cagliari. He is author of several international papers and tutor of many bachelor and master students’ thesis in Electronic Engineering. His research activity is mainly focused on the development of advanced digital systems, with emphasis on the application-driven design and programming of SoCs and FPGAs.
Lecture link:
Course H02-1: Generative AI to gain data insights from the shopfloor intuitively. Aparna Saisree Thuluva, Siemens, Germany.
Course description:
We present a SmartEdge toolkit with several extensions for Mendix, designed to enable creation of Low-Code Edge intelligence apps. First, Mendix connectors for different protocols is presented, e.g., OPC UA, W3C Web of Things (WoT), Bluetooth Low Energy (BLE), and Data Distribution Service (DDS). This facilitates seamless integration of data from different protocols. The “Mendix App Engineering” extension supports the development of industrial IoT solutions using Large Language Models (LLMs) and Low-Code development. Integration with LLMs in Mendix simplifies the discovery of required shop floor data points for engineering by providing an intuitive natural language interface in Mendix. The “Industrial Metaverse Toolchain” simplifies the testing and validation of Mendix applications in virtual environments and their deployment on real-world IoT and automation devices. Lastly, the “ML Toolchain” ensures efficient discovery of Machine Learning tasks and their deployment directly on Edge devices. This solution is implemented and demonstrated on a smart factory use case.
Lecturer(s): Aparna Saisree Thuluva, Siemens, Germany.
Dr. Aparna Thuluva is a senior key expert in the field of IT-OT data integration at Siemens AG, Munich. She focuses on enriching the OT data with semantic and contextual information in order efficiently use the OT data on the IT layer for efficient data insights using AI and Semantic Web technologies.
Lecture link
16:45-17:30
Course G02-2: Computation-in-memory architecture for edge AI applications. Rajendra Bishnoi, Assistant Professor, Delft University of Technology, Netherlands.
Course description:
Neural networks are the foundation of modern artificial intelligence (AI), playing a crucial role in a wide range of cognitive functions, including image classification, voice recognition, self-driving technologies, language understanding, data-driven forecasting, etc. Deploying neural networks on traditional von Neumann architecture hardware, such as CPUs, GPUs, and TPUs, encounters major issues, including poor energy-efficiency and increased latency, primarily due to the memory-wall constraint. Computation-In-Memory (CIM) has the potential to address this problem by integrating processing elements directly into the memory, reducing data movement and enhancing the overall efficiency of the system. The CIM architecture, combined with emerging memristor technologies, delivers various advantages, including leakage-free storage, high density, high scalability, and faster accesses. In this lecture, we will present a comprehensive overview of the memristor-based CIM architecture, including its operational flow and key design principles. We will provide an in-depth analysis of how memristors are utilized to enhance computation efficiency. Additionally, we will explore the intricacies of the hardware- software co-design approach specifically optimized for neural network applications.
Lecturer(s): Rajendra Bishnoi, Assistant Professor, Delft University of Technology, Netherlands.
Rajendra Bishnoi is currently an Assistant Professor with the Computer Engineering Group, Faculty of EEMCS, Delft University of Technology (TU-Delft). He received the Ph.D. degree in computer science from the Karlsruhe Institute of Technology (KIT), Karlsruhe, Germany, in 2017. He was a Research Leader with the MRAM Group, Chair of Dependable Nano Computing, KIT, for more than two years. From 2006 to 2012, he was a Design Engineer with Freescale (now NXP), where he was a part of the Technical Solution Group in memory and SoC flow. His current research interests include neuromorphic computing, computation-in-memory, and emerging technologies. He was a recipient of the EDAA Outstanding Dissertation Award in 2017. He has served on the TPC for several prominent conferences, including DAC’25, DATE’25, DATE’24, ICCAD’24, ETS’24, DATE’23, and DATE’22. In addition, he has held the role of Topic Track Chair for DATE’25, DATE’24, ETS’24, and DSD’23. He is also an editorial board member for the Elsevier Journal and was a guest editor for the Frontiers in Neuroscience Journal in 2022.
Lecture link:
Course H02-2: The role of IoT in edge AI. Jānis Judvaitis, Senior Researcher, Institute of Electronics and Computer Science, Latvia.
Course description:
This course provides an introduction on how the Internet of Things serves as the foundation for Edge AI. The course will explore IoT architectures, sensor networks, data communication protocols and real-time data processing that enable intelligent decision-making at the edge. Overview of lessons learned will be provided from different use cases. Additionally, topics such as IoT security, scalability, and future trends in Edge AI-powered IoT ecosystems will be discussed. Keywords: IoT; edge AI; real time data processing.
Lecturer(s): Jānis Judvaitis, Senior Researcher, Institute of Electronics and Computer Science, Latvia.
Jānis is a senior researcher in Institute of Electronics and Computer Science (EDI) where he has worked from 2013. He is involved in various international projects related to Wireless Sensor Networks, embedded systems, Internet of Things, and edge computing. Expertise in ICT system design, development, implementation, and testing. Author of 20+ scientific articles. Janis defended his doctoral thesis “Enabling Wireless Sensor Network testbeds for high TRL research” in 2023 at the University of Latvia. Research interests include Cyber-Physical Systems, Wireless Sensor Networks, Internet of Things, edge computing and embedded systems.
Lecture link
17:30
End