Progressive Collaboration Pruning for Federated Learning
Collaboration-based personalized federated learning (Co-PFL) mitigates data heterogeneity by enabling clients to selectively exchange knowledge through a learned collaboration graph. Despite their success, existing Co-PFL methods often update collaboration weights mainly from pairwise similarity, which can produce grap...
Siyao Cheng, Boyi Liu, Zimu Zhou et al.
· ACM International Conference... · 0 citations