Bayesian Federated Neural Matching that Completes Full Information

TL;DR

This paper introduces a Bayesian federated neural matching method that incorporates a KL divergence penalty to address information omission, improving model aggregation in federated learning for image classification and segmentation.

Contribution

It proposes a novel Bayesian federated neural matching approach with a KL divergence penalty to enhance information sharing during model aggregation.

Findings

Improved accuracy in image classification tasks.

Enhanced segmentation performance.

Effective model matching in federated learning.

Abstract

Federated learning is a contemporary machine learning paradigm where locally trained models are distilled into a global model. Due to the intrinsic permutation invariance of neural networks, Probabilistic Federated Neural Matching (PFNM) employs a Bayesian nonparametric framework in the generation process of local neurons, and then creates a linear sum assignment formulation in each alternative optimization iteration. But according to our theoretical analysis, the optimization iteration in PFNM omits global information from existing. In this study, we propose a novel approach that overcomes this flaw by introducing a Kullback-Leibler divergence penalty at each iteration. The effectiveness of our approach is demonstrated by experiments on both image classification and semantic segmentation tasks.