L1-Norm Batch Normalization for Efficient Training of Deep Neural Networks

TL;DR

This paper introduces L1-Norm Batch Normalization (L1BN), a linear operation-based normalization method that accelerates training, reduces power consumption, and facilitates low-bit quantization in deep neural networks.

Contribution

L1BN replaces nonlinear operations in batch normalization with linear ones, maintaining accuracy while improving computational efficiency and hardware friendliness.

Findings

L1BN achieves similar accuracy and convergence as L2BN.

On FPGA, L1BN speeds up training by 1.5x and reduces power consumption by 50%.

L1BN enables fully quantized DNN training for future hardware deployment.

Abstract

Batch Normalization (BN) has been proven to be quite effective at accelerating and improving the training of deep neural networks (DNNs). However, BN brings additional computation, consumes more memory and generally slows down the training process by a large margin, which aggravates the training effort. Furthermore, the nonlinear square and root operations in BN also impede the low bit-width quantization techniques, which draws much attention in deep learning hardware community. In this work, we propose an L1-norm BN (L1BN) with only linear operations in both the forward and the backward propagations during training. L1BN is shown to be approximately equivalent to the original L2-norm BN (L2BN) by multiplying a scaling factor. Experiments on various convolutional neural networks (CNNs) and generative adversarial networks (GANs) reveal that L1BN maintains almost the same accuracies and convergence rates compared to L2BN but with higher computational efficiency. On FPGA platform, the proposed signum and absolute operations in L1BN can achieve 1.5$\times$ speedup and save 50\% power consumption, compared with the original costly square and root operations, respectively. This hardware-friendly normalization method not only surpasses L2BN in speed, but also simplify the hardware design of ASIC accelerators with higher energy efficiency. Last but not the least, L1BN promises a fully quantized training of DNNs, which is crucial to future adaptive terminal devices.