English (United Kingdom)

https://curated-unify.zendy.io/wp-json/zendy-region/v1/featured_content/oa?rat=en

https://curated-unify.zendy.io/wp-json/zendy-region/v1/highlighted_journal/

Zendy Plus

Presents the access of premium content as premium feature

Premium Content

Presents the keyphrase highlighting as premium feature

Keyphrase Highlighting

Presents the summarisation as premium feature

Summarisation

Insights

Presents the pdf analysis as premium feature

PDF Analysis

Presents the zaia usage as premium feature

ZAIA

Zendy Tools

Zendy Open

Personalized Federated Learning (pFL) has emerged as a promising solution totackle data heterogeneity across clients in FL. However, existing pFL methodseither (1) introduce high communication and computation costs or (2) overfit tolocal data, which can be limited in scope, and are vulnerable to evolved testsamples with natural shifts. In this paper, we propose PerAda, aparameter-efficient pFL framework that reduces communication and computationalcosts and exhibits superior generalization performance, especially undertest-time distribution shifts. PerAda reduces the costs by leveraging the powerof pretrained models and only updates and communicates a small number ofadditional parameters from adapters. PerAda has good generalization since itregularizes each client's personalized adapter with a global adapter, while theglobal adapter uses knowledge distillation to aggregate generalized informationfrom all clients. Theoretically, we provide generalization bounds to explainwhy PerAda improves generalization, and we prove its convergence to stationarypoints under non-convex settings. Empirically, PerAda demonstrates competitivepersonalized performance (+4.85% on CheXpert) and enables betterout-of-distribution generalization (+5.23% on CIFAR-10-C) on different datasetsacross natural and medical domains compared with baselines, while only updating12.6% of parameters per model based on the adapter. Our code is available athttps://github.com/NVlabs/PerAda.

PerAda: Parameter-Efficient Federated Learning Personalization with Generalization Guarantees