Federated Unsupervised Visual Representation Learning via Exploiting General Content and Personal Style

TL;DR

This paper introduces FedStyle, a federated learning method that leverages local style and content information to improve both global generalization and local personalization of visual representations without labeled data.

Contribution

FedStyle is a novel federated learning approach that exploits style information for unsupervised contrastive learning, enhancing both global generalization and local personalization in decentralized settings.

Findings

FedStyle outperforms baseline methods in both IID and non-IID data settings.

Style infusion and stylized personalization significantly improve representation quality.

Extensive experiments validate the effectiveness of the proposed method.

Abstract

Discriminative unsupervised learning methods such as contrastive learning have demonstrated the ability to learn generalized visual representations on centralized data. It is nonetheless challenging to adapt such methods to a distributed system with unlabeled, private, and heterogeneous client data due to user styles and preferences. Federated learning enables multiple clients to collectively learn a global model without provoking any privacy breach between local clients. On the other hand, another direction of federated learning studies personalized methods to address the local heterogeneity. However, work on solving both generalization and personalization without labels in a decentralized setting remains unfamiliar. In this work, we propose a novel method, FedStyle, to learn a more generalized global model by infusing local style information with local content information for contrastive learning, and to learn more personalized local models by inducing local style information for downstream tasks. The style information is extracted by contrasting original local data with strongly augmented local data (Sobel filtered images). Through extensive experiments with linear evaluations in both IID and non-IID settings, we demonstrate that FedStyle outperforms both the generalization baseline methods and personalization baseline methods in a stylized decentralized setting. Through comprehensive ablations, we demonstrate our design of style infusion and stylized personalization improve performance significantly.