Compare Where It Matters: Using Layer-Wise Regularization To Improve Federated Learning on Heterogeneous Data

TL;DR

This study shows that only certain important layers in a neural network require regularization for effective training, and presents FedCKA, a simple framework that outperforms previous state-of-the-art methods on various deep learning tasks while also improving efficiency and scalability.

Abstract

Federated Learning is a widely adopted method to train neural networks over distributed data. One main limitation is the performance degradation that occurs when data is heterogeneously distributed. While many works have attempted to address this problem, these methods under-perform because they are founded on a limited understanding of neural networks. In this work, we verify that only certain important layers in a neural network require regularization for effective training. We additionally verify that Centered Kernel Alignment (CKA) most accurately calculates similarity between layers of neural networks trained on different data. By applying CKA-based regularization to important layers during training, we significantly improve performance in heterogeneous settings. We present FedCKA: a simple framework that out-performs previous state-of-the-art methods on various deep learning tasks while also improving efficiency and scalability.

Figures (5)

• Figure 1: The typical steps of Federated Learning
• Figure 2: CKA similarity comparison between each client and the global model (Refer to Experimental Setup for more information on setup)
• Figure 3: Training Process of FedCKA
• Figure 4: Distribution of the CIFAR-10 dataset across 10 clients according to the Dirichlet distribution. The x-axis shows the index of the client, and the y-axis shows the index of the class (label). As the parameter $\alpha$ approaches 0, there the heterogeneity of class distribution increases.
• Figure 5: Accuracy with respect to the number of layers regularized on CIFAR-10 and $\alpha = 5.0$