CRIP: Channel Level Representation Injection for Personalized One-Shot Federated Learning

2026-08-03

AI Summary

This work addresses the challenges of feature shift and insufficient personalization in single-round federated learning under severe domain heterogeneity, where the absence of iterative interaction limits adaptation. To overcome this, the authors propose a personalized framework based on channel-level feature alignment. Their approach introduces, for the first time in a single-round setting, a channel-wise representation similarity metric that enables selective fusion of features from source clients most compatible with the target client—without requiring auxiliary data—thereby achieving fine-grained and domain-noise-robust knowledge transfer. By integrating channel alignment, local mini-batch similarity evaluation, and selective fusion, the method significantly outperforms both local training and existing state-of-the-art approaches on highly heterogeneous benchmarks including DomainNet, PACS, and Office-Home, demonstrating the efficacy of representation-space personalization under extreme domain shifts.

Abstract

One-shot federated learning (OSFL) has emerged as a promising collaborative model learning framework with only a single round of communication, offering significant advantages in communication efficiency and privacy preservation. However, OSFL often faces inherent limitations under severe domain heterogeneity across clients due to the lack of iterative knowledge exchange. Most existing OSFL methods require an auxiliary public dataset for knowledge distillation or leverage statistical information for parameter-level aggregation, overlooking feature shift caused by domain heterogeneity. To address these challenges, we propose CRIP, a personalized OSFL framework that operates in the representation space via channel-level feature alignment. To achieve this, each client uploads its feature extractor to the server, which broadcasts all extractors back to every client. Since not all source clients share compatible feature distributions with the target client, indiscriminate fusion of cross-client features would introduce domain-specific noise. Therefore, CRIP effectively measures the channel-wise representational similarity between the target client and each source client on a small local mini-batch, and selectively fuses only the most compatible features. Extensive experiments on domain-heterogeneous benchmarks such as DomainNet, PACS, and Office-Home demonstrate that CRIP consistently outperforms local models and state-of-the-art baselines, validating the effectiveness of representation-space personalization under extreme domain heterogeneity.

Problem

• one-shot federated learning
• domain heterogeneity
• feature shift
• personalization
• representation space

Innovation

• one-shot feder日晚间 learning
• representation alignment
• channel-level fusion
• domain heterogeneity
• personalized federated learning