Federated learning enables multiple participants to collaboratively train machine learning models while keeping their datasets private. However, in untrusted environments, ensuring both privacy and robustness to malicious participants remains a fundamental challenge, particularly in the design of secure aggregation mechanisms. In this talk, we give an overview of recent developments in secure aggregation methods for federated learning, and then focus on a recent approach called FORTA, which leverages analog error-correction codes together with Krum-based outlier detection to achieve both privacy and robustness. We will discuss the key ideas behind the framework along with supporting theoretical and experimental results.
Prof. Harshan Jagadeesh is an associate professor with the Department of Electrical Engineering, IIT Delhi. His research interests lie in the broad area of network security, with a special focus on applying information security techniques to communication, storage, and computing systems.