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 Plus annual

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

Global: Zendy Plus monthly

Global: Zendy Tools annual

Global: Zendy Tools monthly

Global: Zendy Free Plan

Federated learning (FL) is a privacy-preserving distributed learning paradigmthat enables clients to jointly train a global model. In real-world FLimplementations, client data could have label noise, and different clientscould have vastly different label noise levels. Although there exist methods incentralized learning for tackling label noise, such methods do not perform wellon heterogeneous label noise in FL settings, due to the typically smaller sizesof client datasets and data privacy requirements in FL. In this paper, wepropose $\texttt{FedCorr}$, a general multi-stage framework to tackleheterogeneous label noise in FL, without making any assumptions on the noisemodels of local clients, while still maintaining client data privacy. Inparticular, (1) $\texttt{FedCorr}$ dynamically identifies noisy clients byexploiting the dimensionalities of the model prediction subspaces independentlymeasured on all clients, and then identifies incorrect labels on noisy clientsbased on per-sample losses. To deal with data heterogeneity and to increasetraining stability, we propose an adaptive local proximal regularization termthat is based on estimated local noise levels. (2) We further finetune theglobal model on identified clean clients and correct the noisy labels for theremaining noisy clients after finetuning. (3) Finally, we apply the usualtraining on all clients to make full use of all local data. Experimentsconducted on CIFAR-10/100 with federated synthetic label noise, and on areal-world noisy dataset, Clothing1M, demonstrate that $\texttt{FedCorr}$ isrobust to label noise and substantially outperforms the state-of-the-artmethods at multiple noise levels.

FedCorr: Multi-Stage Federated Learning for Label Noise Correction