Contrast with major classifier vectors for federated medical relation extraction with heterogeneous label distribution.

Federated medical relation extraction enables multiple clients to train a deep network collaboratively without sharing their raw medical data. To handle the heterogeneous label distribution across clients, most of the existing works enforce regularization between local and global models during updating. In this paper, we propose the concept of major classifier vectors, which are a group of classifier vectors that characterize the representation space of relation classes well. They are obtained by comparing the inter-classifier similarity between clients, which is an ensembling method that avoids the bias introduced by weighted aggregation. We propose an algorithm named FedCMC, which restricts the updating of local models by contrasting with major classifier vectors to avoid overfitting to the local label distribution by comparison with major classifier vectors. Extensive experiments show that FedCMC outperforms the other state-of-the-art federated learning (FL) algorithms on three medical relation extraction datasets. [ABSTRACT FROM AUTHOR]