FedGKD: Toward Heterogeneous Federated Learning via Global Knowledge Distillation

Abstract

Federated learning, as one enabling technology of edge intelligence, has gained substantial attention due to its efficacy in training deep learning models without data privacy and network bandwidth concerns. However, due to the heterogeneity of the edge computing system and data, many methods suffer from the <italic xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">“client-drift”</i> issue that could considerably impede the convergence of global model training: local models on clients can drift apart, and the aggregated model can be different from the global optimum. To tackle this issue, one intuitive idea is to guide the local model training by global teachers, <italic xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">i.e.,</i> past global models, where each client learns the global knowledge from past global models via adaptive knowledge distillation techniques. Inspired by these insights, we propose a novel approach for heterogeneous federated learning, <sc xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">FedGKD</small> , which fuses the knowledge from historical global models and guides local training to alleviate the <italic xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">“client-drift”</i> issue. In this paper, we evaluate <sc xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">FedGKD</small> through extensive experiments across various CV and NLP datasets ( <italic xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">i.e.,</i> CIFAR-10/100, Tiny-ImageNet, AG News, SST5) under different heterogeneous settings. The proposed method is guaranteed to converge under common assumptions and outperforms the state-of-the-art baselines in the non-IID federated setting.

Topics & Concepts