Search papers, labs, and topics across Lattice.
This paper introduces CRIP, a personalized one-shot federated learning framework that enhances model performance by aligning channel-level features across clients facing severe domain heterogeneity. By allowing clients to upload their feature extractors and selectively fusing compatible features based on representational similarity, CRIP mitigates the noise introduced by incompatible feature distributions. Extensive experiments on benchmark datasets like DomainNet and PACS show that CRIP significantly outperforms existing local models and state-of-the-art methods, demonstrating its effectiveness in challenging federated learning scenarios.
CRIP selectively fuses features from clients based on representational similarity, leading to superior performance in one-shot federated learning under extreme domain heterogeneity.
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.
