TrajSyn: Privacy-Preserving Dataset Distillation from Federated Model Trajectories for Server-Side Adversarial Training

TL;DR

TrajSyn is a privacy-preserving method that synthesizes datasets from federated model updates to enable effective server-side adversarial training without compromising client data privacy.

Contribution

It introduces a novel framework that creates proxy datasets from model trajectories, facilitating robust adversarial training in federated learning settings.

Findings

Improves adversarial robustness on image classification benchmarks.

No additional compute burden on client devices.

Maintains client data privacy while enabling effective training.

Abstract

Deep learning models deployed on edge devices are increasingly used in safety-critical applications. However, their vulnerability to adversarial perturbations poses significant risks, especially in Federated Learning (FL) settings where identical models are distributed across thousands of clients. While adversarial training is a strong defense, it is difficult to apply in FL due to strict client-data privacy constraints and the limited compute available on edge devices. In this work, we introduce TrajSyn, a privacy-preserving framework that enables effective server-side adversarial training by synthesizing a proxy dataset from the trajectories of client model updates, without accessing raw client data. We show that TrajSyn consistently improves adversarial robustness on image classification benchmarks with no extra compute burden on the client device.