Personalized Federated Learning

Abstract:
Federated Learning (FL) has emerged as a prominent methodology for collectively training a shared machine learning model over the Internet while maintaining data at client endpoints. However, due to variations in data distributions among clients, the efficacy of the shared model may be limited for certain clients, prompting the exploration of FL techniques aimed at training personalized models.
This talk will provide a succinct overview of three recent FL approaches, each grounded in distinct assumptions and accompanied by corresponding learning algorithms: (1) kNNPer, which leverages a shared representation; (2) FedEM, which assumes local client distributions as mixtures of a common set of underlying distributions; and (3) ColME, which operates under the assumption of shared distributions across clusters of clients.

References:
- Marfoq, O., Neglia, G., Vidal, R., & Kameni, L. (2022). Personalized Federated Learning through Local Memorization. In Proceedings of the 39th International Conference on Machine Learning (PMLR 162:15070-15092). Available at:https://proceedings.mlr.press/v162/marfoq22a.html.
- Marfoq, O., Neglia, G., Bellet, A., Kameni, L., & Vidal, R. (2021). Federated Multi-Task Learning under a Mixture of Distributions. Advances in Neural Information Processing Systems 34 (NeurIPS 2021). Available at:https://proceedings.neurips.cc/paper/2021/hash/82599a4ec94aca066873c99b4c741ed8-Abstract.html.
- Galante, F., Neglia, G., & Leonardi, E. (2024). Scalable Decentralized Algorithms for Online Personalized Mean Estimation. Soon to be available on arxiv.

Bio:
Dr. Giovanni Neglia is a researcher at Inria, France, since 2008, and holds a chair on Pervasive Sustainable Learning Systems at the 3IA Côte d'Azur (one of the French Interdisciplinary Institutes on Artificial Intelligence), since 2021. He received his Habilitation in 2017 from the Université Côte d'Azur, France, his PhD and electronic engineering degree from University of Palermo, Italy, respectively in 2005 and in 2001. Before joining Inria as a permanent researcher, he was a research scholar at the University Massachusetts Amherst, (2005) and a postdoc at Inria (2006-2007). His research activity focuses on modelling and performance evaluation of networked systems and proposals of new mechanisms to improve their performance. Currently, his main interests are federated learning and online learning