FedNS: Improving Federated Learning for collaborative image classification on mobile clients

TL;DR

FedNS introduces a novel client model aggregation method in federated learning that filters and re-weights client contributions at the node level, leading to improved image classification performance across datasets.

Contribution

This paper proposes FedNS, a new server-side aggregation approach that enhances federated learning by selectively combining client models at the component level.

Findings

FedNS outperforms FedAvg in multiple image classification datasets.

FedNS achieves higher accuracy and robustness.

The method effectively filters out less useful client contributions.

Abstract

Federated Learning (FL) is a paradigm that aims to support loosely connected clients in learning a global model collaboratively with the help of a centralized server. The most popular FL algorithm is Federated Averaging (FedAvg), which is based on taking weighted average of the client models, with the weights determined largely based on dataset sizes at the clients. In this paper, we propose a new approach, termed Federated Node Selection (FedNS), for the server's global model aggregation in the FL setting. FedNS filters and re-weights the clients' models at the node/kernel level, hence leading to a potentially better global model by fusing the best components of the clients. Using collaborative image classification as an example, we show with experiments from multiple datasets and networks that FedNS can consistently achieve improved performance over FedAvg.