How Does Batch Normalization Help Binary Training?

Abstract

Binary Neural Networks (BNNs) are difficult to train, and suffer from drop of accuracy. It appears in practice that BNNs fail to train in the absence of Batch Normalization (BatchNorm) layer. We find the main role of BatchNorm is to avoid exploding gradients in the case of BNNs. This finding suggests that the common initialization methods developed for full-precision networks are irrelevant to BNNs. We build a theoretical study on the role of BatchNorm in binary training, backed up by numerical experiments.

References in corpus (4)

• MobileNets: Efficient Convolutional Neural Networks for Mobile Vision Applications
• Communication-Efficient Learning of Deep Networks from Decentralized Data
• Deep Residual Learning for Image Recognition
• Quantized Neural Networks: Training Neural Networks with Low Precision Weights and Activations

Cited by in corpus (9)

• Normalization Techniques in Training DNNs: Methodology, Analysis and Application
• Binary Neural Networks for Memory-Efficient and Effective Visual Place Recognition in Changing Environments
• Jet Single Shot Detection
• PR-CIM: a Variation-Aware Binary-Neural-Network Framework for Process-Resilient Computation-in-memory
• Batch Normalization in Quantized Networks
• Bit Error Tolerance Metrics for Binarized Neural Networks
• Enabling Binary Neural Network Training on the Edge
• Universal Approximation Theorems of Fully Connected Binarized Neural Networks
• Random Bias Initialization Improves Quantized Training