Bilkent University
Department of Computer Engineering
CS 590/690 SEMINAR
Exploring Privacy-Preserving Normalization Techniques in Federated Learning
Melih Coşğun
Master Student
(Supervisor: Asst.Prof.Sinem Sav)
Computer Engineering Department
Bilkent University
Abstract: Data normalization is a fundamental preprocessing step in machine learning for improving model performance and training stability. In federated learning (FL), a collaborative machine learning approach where data remains distributed across multiple clients while jointly training a model, normalization poses unique challenges due to the decentralized nature of the data. Traditional methods rely on either independent client-side processing, i.e., local normalization, or normalizing the entire dataset before distributing it to clients, i.e., pooled normalization. Local normalization can be problematic when data distributions across clients are non-IID, while adopting a pooled normalization approach conflicts with the decentralized nature of FL. In this project, we explore the adaptation of widely used normalization techniques to FL and define the term federated normalization. Federated normalization simulates pooled normalization by enabling the collaborative exchange of normalization parameters among clients. Thus, federated normalization achieves performance on par with pooled normalization and it preserves the privacy guarantees of local normalization, as raw data remains securely on the clients' devices.
DATE: March 17, Monday @ 14:30 Place: EA 502