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/

Global: Zendy Tools annual

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

Global: Zendy Tools monthly

Global: Zendy Plus monthly

Presents the access of premium content as premium feature

Premium Content

Global: Zendy Plus annual

Global: Zendy Free Plan

Federated Learning (FL) enables multiple clients to collaboratively learn ina distributed way, allowing for privacy protection. However, the real-worldnon-IID data will lead to client drift which degrades the performance of FL.Interestingly, we find that the difference in logits between the local andglobal models increases as the model is continuously updated, thus seriouslydeteriorating FL performance. This is mainly due to catastrophic forgettingcaused by data heterogeneity between clients. To alleviate this problem, wepropose a new algorithm, named FedCSD, a Class prototype SimilarityDistillation in a federated framework to align the local and global models.FedCSD does not simply transfer global knowledge to local clients, as anundertrained global model cannot provide reliable knowledge, i.e., classsimilarity information, and its wrong soft labels will mislead the optimizationof local models. Concretely, FedCSD introduces a class prototype similaritydistillation to align the local logits with the refined global logits that areweighted by the similarity between local logits and the global prototype. Toenhance the quality of global logits, FedCSD adopts an adaptive mask to filterout the terrible soft labels of the global models, thereby preventing them tomislead local optimization. Extensive experiments demonstrate the superiorityof our method over the state-of-the-art federated learning approaches invarious heterogeneous settings. The source code will be released.

Rethinking Client Drift in Federated Learning: A Logit Perspective