FedBN: Federated Learning on Non-IID Features via Local Batch Normalization

Abstract

The emerging paradigm of federated learning (FL) strives to enable collaborative training of deep models on the network edge without centrally aggregating raw data and hence improving data privacy. In most cases, the assumption of independent and identically distributed samples across local clients does not hold for federated learning setups. Under this setting, neural network training performance may vary significantly according to the data distribution and even hurt training convergence. Most of the previous work has focused on a difference in the distribution of labels or client shifts. Unlike those settings, we address an important problem of FL, e.g., different scanners/sensors in medical imaging, different scenery distribution in autonomous driving (highway vs. city), where local clients store examples with different distributions compared to other clients, which we denote as feature shift non-iid. In this work, we propose an effective method that uses local batch normalization to alleviate the feature shift before averaging models. The resulting scheme, called FedBN, outperforms both classical FedAvg, as well as the state-of-the-art for non-iid data (FedProx) on our extensive experiments. These empirical results are supported by a convergence analysis that shows in a simplified setting that FedBN has a faster convergence rate than FedAvg. Code is available at https://github.com/med-air/FedBN.

Accepted at ICLR 2021

References in corpus (15)

• Communication-Efficient Learning of Deep Networks from Decentralized Data
• Federated Learning with Non-IID Data
• Neural Tangent Kernel: Convergence and Generalization in Neural Networks
• On the Convergence of FedAvg on Non-IID Data
• A Convergence Theory for Deep Learning via Over-Parameterization
• Measuring the Effects of Non-Identical Data Distribution for Federated Visual Classification
• Revisiting Batch Normalization For Practical Domain Adaptation
• Gradient Descent Provably Optimizes Over-parameterized Neural Networks
• FedML: A Research Library and Benchmark for Federated Machine Learning
• A generic framework for privacy preserving deep learning
• Fine-Grained Analysis of Optimization and Generalization for Overparameterized Two-Layer Neural Networks
• On the importance of single directions for generalization
• Adaptive Federated Optimization
• Flower: A Friendly Federated Learning Research Framework
• The Non-IID Data Quagmire of Decentralized Machine Learning

Cited by in corpus (17)

• Do Gradient Inversion Attacks Make Federated Learning Unsafe?
• Memory-aware curriculum federated learning for breast cancer classification
• Federated Multi-Task Learning under a Mixture of Distributions
• Federated Learning on Non-IID Data Silos: An Experimental Study
• SemiFL: Semi-Supervised Federated Learning for Unlabeled Clients with Alternate Training
• Quasi-Global Momentum: Accelerating Decentralized Deep Learning on Heterogeneous Data
• Federated Reconstruction: Partially Local Federated Learning
• Connecting Low-Loss Subspace for Personalized Federated Learning
• Federated Learning for Smart Healthcare: A Survey
• You Can Backdoor Personalized Federated Learning
• FedDR -- Randomized Douglas-Rachford Splitting Algorithms for Nonconvex Federated Composite Optimization
• FL-NTK: A Neural Tangent Kernel-based Framework for Federated Learning Convergence Analysis
• Gradient Inversion with Generative Image Prior
• Personalized Retrogress-Resilient Framework for Real-World Medical Federated Learning
• Neural Tangent Kernel Empowered Federated Learning
• DVFL: A Vertical Federated Learning Method for Dynamic Data
• Federated Learning from Small Datasets