Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

Feature Diversification and Adaptation for Federated Domain Generalization

Seunghan Yang, Seokeon Choi, Hyunsin Park, Sungha Choi, Simyung Chang, Sungrack Yun

TL;DR

The paper tackles federated domain generalization under privacy constraints, where client data come from distinct source domains and visible distribution shifts hinder unseen-domain performance. It introduces FedFD-A, which combines federated feature diversification—leveraging aggregated global feature statistics to diversify local data and foster client-invariant representations—with an instance feature adapter for test-time alignment using both local and global statistics. The approach preserves privacy by avoiding raw data sharing while enabling robust generalization on benchmarks such as PACS, VLCS, OfficeHome, and DomainNet, achieving state-of-the-art performance in federated DG. Overall, FedFD-A provides a practical, privacy-preserving solution for cross-domain generalization in distributed learning settings, with strong potential for real-world deployment.

Abstract

Federated learning, a distributed learning paradigm, utilizes multiple clients to build a robust global model. In real-world applications, local clients often operate within their limited domains, leading to a `domain shift' across clients. Privacy concerns limit each client's learning to its own domain data, which increase the risk of overfitting. Moreover, the process of aggregating models trained on own limited domain can be potentially lead to a significant degradation in the global model performance. To deal with these challenges, we introduce the concept of federated feature diversification. Each client diversifies the own limited domain data by leveraging global feature statistics, i.e., the aggregated average statistics over all participating clients, shared through the global model's parameters. This data diversification helps local models to learn client-invariant representations while preserving privacy. Our resultant global model shows robust performance on unseen test domain data. To enhance performance further, we develop an instance-adaptive inference approach tailored for test domain data. Our proposed instance feature adapter dynamically adjusts feature statistics to align with the test input, thereby reducing the domain gap between the test and training domains. We show that our method achieves state-of-the-art performance on several domain generalization benchmarks within a federated learning setting.

Feature Diversification and Adaptation for Federated Domain Generalization

TL;DR

The paper tackles federated domain generalization under privacy constraints, where client data come from distinct source domains and visible distribution shifts hinder unseen-domain performance. It introduces FedFD-A, which combines federated feature diversification—leveraging aggregated global feature statistics to diversify local data and foster client-invariant representations—with an instance feature adapter for test-time alignment using both local and global statistics. The approach preserves privacy by avoiding raw data sharing while enabling robust generalization on benchmarks such as PACS, VLCS, OfficeHome, and DomainNet, achieving state-of-the-art performance in federated DG. Overall, FedFD-A provides a practical, privacy-preserving solution for cross-domain generalization in distributed learning settings, with strong potential for real-world deployment.

Abstract

Federated learning, a distributed learning paradigm, utilizes multiple clients to build a robust global model. In real-world applications, local clients often operate within their limited domains, leading to a `domain shift' across clients. Privacy concerns limit each client's learning to its own domain data, which increase the risk of overfitting. Moreover, the process of aggregating models trained on own limited domain can be potentially lead to a significant degradation in the global model performance. To deal with these challenges, we introduce the concept of federated feature diversification. Each client diversifies the own limited domain data by leveraging global feature statistics, i.e., the aggregated average statistics over all participating clients, shared through the global model's parameters. This data diversification helps local models to learn client-invariant representations while preserving privacy. Our resultant global model shows robust performance on unseen test domain data. To enhance performance further, we develop an instance-adaptive inference approach tailored for test domain data. Our proposed instance feature adapter dynamically adjusts feature statistics to align with the test input, thereby reducing the domain gap between the test and training domains. We show that our method achieves state-of-the-art performance on several domain generalization benchmarks within a federated learning setting.
Paper Structure (6 sections, 2 figures)

This paper contains 6 sections, 2 figures.

Table of Contents

  1. Introduction
  2. Related Works
  3. Federated Learning
  4. Domain Generalization
  5. Proposed Methods
  6. Preliminaries

Figures (2)

  • Figure 1: Our Concept Diagram. Federated feature diversification: during training, local data is diversified using global feature statistics—averaged statistics across all client data—provided by the global model. This approach aids in learning representations that are invariant to individual client characteristics. Instance feature adaptation: in the test phase, the adapter employs input feature statistics to facilitate instance-adaptive inference tailored for new and unseen test data.
  • Figure :