Understanding and Improving Model Averaging in Federated Learning on Heterogeneous Data
Tailin Zhou, Zehong Lin, Jun Zhang, Danny H. K. Tsang
TL;DR
Federated learning with heterogeneous data exhibits strong empirical performance from model averaging, but the mechanics were not well understood. The authors visualize loss landscapes to reveal that the global model often lies within a basin formed by client models yet can deviate from the basin center, and they decompose the global loss into five factors (TrainBias, HeterBias, Var, Cov, Locality) to explain this behavior. They connect FMA to a weighted ensemble of outputs (WENS) and propose Iterative Moving Averaging (IMA) with mild client exploration to keep the global model near the basin center during late training, improving accuracy and speed across benchmarks. Across diverse heterogeneous setups, IMA yields consistent gains and reduced communication overhead, providing a geometry-informed, practical enhancement to FL in non-iid environments.
Abstract
Model averaging is a widely adopted technique in federated learning (FL) that aggregates multiple client models to obtain a global model. Remarkably, model averaging in FL yields a superior global model, even when client models are trained with non-convex objective functions and on heterogeneous local datasets. However, the rationale behind its success remains poorly understood. To shed light on this issue, we first visualize the loss landscape of FL over client and global models to illustrate their geometric properties. The visualization shows that the client models encompass the global model within a common basin, and interestingly, the global model may deviate from the basin's center while still outperforming the client models. To gain further insights into model averaging in FL, we decompose the expected loss of the global model into five factors related to the client models. Specifically, our analysis reveals that the global model loss after early training mainly arises from \textit{i)} the client model's loss on non-overlapping data between client datasets and the global dataset and \textit{ii)} the maximum distance between the global and client models. Based on the findings from our loss landscape visualization and loss decomposition, we propose utilizing iterative moving averaging (IMA) on the global model at the late training phase to reduce its deviation from the expected minimum, while constraining client exploration to limit the maximum distance between the global and client models. Our experiments demonstrate that incorporating IMA into existing FL methods significantly improves their accuracy and training speed on various heterogeneous data setups of benchmark datasets. Code is available at \url{https://github.com/TailinZhou/FedIMA}.