Machine Learning for Edge-Cloud Systems (ML4ECS)
CfP

The second edition of the Machine Learning for Edge-Cloud Systems (ML4ECS) workshop will be jointly organized by three Horizon Europe projects: CODECO, Edgeless, and MLSysOps. This collaborative event will highlight cutting-edge research and innovation in ML-based, autonomic end-to-end system management and application deployment and execution across the IoT-Edge-Cloud continuum—an area of growing importance due to the increasing complexity of such systems.

The workshop focuses on the research outcomes of the three organizing projects. Key topics include machine learning (ML) models to improve efficiency, reduce latency, and enhance performance across the edge-cloud continuum, integration with edge and cloud systems, optimization, security, and privacy-preserving decentralized learning.

The workshop will feature:

• Technical sessions showcasing the latest research outcomes and practical use cases from the three organizing projects,
• A keynote speech by a prominent expert in the field, and
• An interactive discussion panel addressing future challenges and opportunities in ML-driven optimizations of system functionality across IoT-Edge-Cloud environments.

The ML4ECS workshop aims to bridge academic and industrial perspectives, fostering dialogue and collaboration on deploying machine learning models to enable intelligent, adaptive, and efficient system and application management throughout the IoT-Edge-Cloud continuum.

9:30 – 10:45
HiPEAC Conference Welcome and Keynote

11:00 – 12:00
ML4ECS Keynote Speech

Lorenzo Valerio (IIT-CNR)

Title: Decentralized Federated Learning Over Edge Networks: a Coordination-Free Learning Substrate for Agentic AI

Abstract: The rapid emergence of agentic AI systems—where multiple autonomous components sense, decide, and act in a coordinated way—is shifting AI from “train once, deploy everywhere” to continuous learning in the loop: each agent generates experience locally and must improve from it while cooperating with others operating under different conditions. This creates an immediate systems-related question: how do agents share what they learn without moving raw observations to a central hub, and without relying on a single coordinator that becomes a bottleneck or a failure domain? In edge and IoT settings the problem is amplified by heterogeneity: data and rewards are unevenly distributed across nodes, devices and links have fluctuating capacity that prevents stable participation, and partial connectivity and community structure mean that who can talk to whom strongly shapes what knowledge can actually propagate.
Decentralized Federated Learning (DFL) provides a concrete answer to this coordination problem by replacing server-centric aggregation with peer-to-peer model exchange and gossip-style mixing, turning the communication topology into a first-class actor, not always fully controllable (in contrast to the full controllability of datacenters). In this view, DFL is the learning substrate for agentic systems: it enables agents to pool experience into shared models/policies (or multiple community-specific ones) while remaining robust to churn, partitions, and bandwidth constraints typical of edge networks. This keynote will introduce the interplay between agentic AI and DFL from a networking and distributed-systems perspective, reviewing recent advances and deriving what DFL can already deliver for agentic deployments—scalable coordination, resilience, and topology-aware “knowledge spreading”—and what remains open.

Short bio: Dr. Lorenzo Valerio is a Senior Researcher at IIT-CNR. His research focuses on decentralized and resource-constrained machine learning for Edge/IoT environments, with interests spanning decentralized federated learning, deep learning, causal learning, generative models, and opportunistic/mobile networking. Since 2021, he has co-organized the IEEE PeRConAI Workshop and has served as Guest Editor for several international journals. He has received three Best Paper Awards and one Best Paper Nomination. He currently serves on the Editorial Board of Elsevier Computer Communications and regularly contributes to the technical program committees of leading conferences, (including AAAI, IEEE PerCom, IEEE IJCNN). He is (or has been) involved in multiple European and national research projects, spanning Foundational AI as well as AI applied to several domains (e.g., terrestrial and non terrestrial networks, big data analysis, Smart Cities, Industry 4.0 and more).

12:00 – 12:30
Coffee Break

12:30 – 14:00
EDGELESS

• 12:30 – 12:45: Manolis Marazakis (FORTH, Greece). Edge and Cloud Systems Based on Open Standards.
• 12:45 – 13:00: Ioannis Morianos (Dienekes, Greece). Hardware (FPGA) Accelerated Deep Learning for Edge Intrusion Detection Systems.
• 13:00 – 13:15: Claudio Cicconetti (CNR, Italy). Enabling the Stateful FaaS Agent Pattern in EDGELESS.
• 13:15 – 13:30: Jonathan Cacace (Eurecat, Spain) and Francisco Vicente Parra (Worldline, Spain). EDGELESS Videos: Internet of Robotic Things & The Function Repository.
• 13:30 – 13:45: Saurabh Singh (Siemens, Germany). Beyond the Cloud: Industrializing Serverless with FaaS at the Edge.
• 13:45 – 14:00: Bruno Pereira (Ubiwhere, Portugal). EDGELESS: A use case on Smart City Analytics.

14:00 – 15:00
Lunch

15:00 – 16:30
CODECO

• 15:00 – 15:15: Rute C. Sofia, (fortiss, Germany). CODECO Federated Operation Framework.
• 15:15 – 15:30: Giorgios Papathanail (Intracom-Telecom). CODECO OSS Across Multi-Clusters (Demo).
• 15:30 – 15:45: Panagiotis Karamolegkos (University of Piraeus Research Center) and (Alejandro Espinosa, I2CAT). Privacy-Preserving Decentralised Learning (Demo).
• 15:45 – 16:00: George Koukis (ATHENA).CODEF Demo.
• 16:00 – 16:20: Panagiotis Karamolegkos (University of Piraeus Research Center). CODECO Use Cases and real-world applications.

16:30 – 17:00
Coffee Break

17:00 – 18:30
MLSysOps

• 17:00 – 17:05: Spyros Lalis (U Thessaly, Greece). MLSysOps Overview and Architecture.
• 17:05 – 17:20: Bruno Pereira (Ubiwhere, Portugal). Smart cities.
• 17:20 – 17:35: Phivos Papapanagiotakis, Giorgos Giannios (CNH, Greece). Enhancing Targeted Weed Spraying with Drone-Tractor Collaboration. ​
• 17:35 – 17:50: Theodoros Aslanidis, Dimitris Chatzopoulos (UCD, Ireland). AI Pipelines, MLOps, and AutoML in the Continuum.
• 17:50 – 18:05: Marcell Feher (Chocolate Cloud, Denmark). Traffic-Aware Object Storage via ML-Based Dynamic Policy Reconfiguration.
• 18:05 – 18:20: Alexandros Patras (U Thessaly, Greece). Optimizing ML Inference in FPGAs Using Reinforcement Learning.
• 18:20 – 18:35: Anastasios Nanos (NUBIS, Greece). Make your Object Detection Pipeline Smarter with MLSysOps and vAccel.

18:30
End of Workshop

• MLSysOps: Nikolaos Bellas (University of Thessaly, Greece), Raffaele Gravina (University of Calabria, Italy)
• EDGELESS: Claudio Cicconetti (CNR, Italy), Jesús Omaña Iglesias (Telefónica, Spain)
• CODECO: Rute C. SOFIA (fortiss GmbH, Germany)

Technical Program Committee:

TBA