Seminar on March 25, 2024

Title

Resilient federated learning: Where performance meets constraints

Speaker

Dr. Yang Lu, Lancaster University, UK

When and Where

March 25, 2024, 16:00-17:00
Online on Zoom, see TAS Resilience Node for registration or send an email to Valeria Cardellini

Abstract

Federated learning is increasingly employed in multi-agent systems such as cyber-physical systems applications, enhancing network intelligence. However, such applications in turn raise new constraints and challenges to existing frameworks. Critical challenges include, e.g., user mobility, data heterogeneity, sparse communication topology, absence of a central server, open and insecure communication links, and external and internal attackers. In this talk, I will discuss three of my recent works in attempting to achieve high-quality federated learning while tackling these challenges. Specifically, I will cover adaptive federated learning over dynamic and heterogeneous users, robust federated learning against poisoning attacks, and privacy-preserving peer-to-peer federated learning.