2023

Titeln är decentralized adaptive clustering is beneficial for client collaboration.

Edvin Listo Zec, Ebba Ekblom, Martin Willbo, Olof Mogren, Sarunas Girdzijauskas

Summary

Decentralized adaptive clustering of deep nets is beneficial for client collaboration. DAC (Decentralized Adaptive Clustering) helps clients with similar data distributions communicate more frequently using similarity measure based on softmax outputs. Improves accuracy on covariate and label shift scenarios.

Themes

federated-learningmachine-learning

Keywords

decentralized learning, adaptive clustering, federated learning, non-iid data, deep learning

Poster

Click image to open full size