The seminar will be titled “Federated Learning at the Edge: Addressing Data Heterogeneity in IoT Systems”
Federated Learning (FL) and Edge AI are important building blocks of scalable and privacy-preserving intelligence in Internet of Things (IoT)-based systems. However, real-world deployments are inherently affected by data heterogeneity (non-IID distributions) across settings, which significantly degrades model performance and convergence. In this talk, we present a system-oriented approach to edge intelligence, combining on-device inference, federated training, and architecture-level solutions to address heterogeneity. We begin with edge-native AI pipelines for real-time sensing and inference, highlighting how local processing reduces latency and communication overhead. We then discuss federated transfer learning strategies that enable collaborative model training across distributed clients while preserving data locality. Finally, a novel federated architecture based on a client-shared latent space, which improves robustness to non-IID data by aligning semantic representations across clients while reducing communication costs.
The speaker
Ivan Zyrianoff received the B.S. degree in computer science and the M.S. degree in information engineering from the Federal University of ABC, Santo André, Brazil, in 2017 and 2019, respectively, and the Ph.D. degree from the University of Bologna, Bologna, Italy, in 2024. He is a Research Fellow from the University of Bologna, Bologna, Italy, and a member of the IoT-Prism Lab. His current research topics encompass interoperability for the Internet of Things, edge computing and intelligence, and proactive caching.
