Estimator Meets Equilibrium Perspective: A Rectified Straight Through Estimator for Binary Neural Networks Training

TL;DR

This paper introduces ReSTE, a new estimator for binary neural network training that balances estimation accuracy and gradient stability, leading to improved performance without extra modules.

Contribution

It proposes a novel perspective on BNN training as an equilibrium between estimation error and gradient stability, and designs ReSTE to effectively balance these factors.

Findings

ReSTE outperforms existing estimators on CIFAR-10 and ImageNet.

ReSTE achieves state-of-the-art results without auxiliary modules.

The equilibrium perspective guides better estimator design.

Abstract

Binarization of neural networks is a dominant paradigm in neural networks compression. The pioneering work BinaryConnect uses Straight Through Estimator (STE) to mimic the gradients of the sign function, but it also causes the crucial inconsistency problem. Most of the previous methods design different estimators instead of STE to mitigate it. However, they ignore the fact that when reducing the estimating error, the gradient stability will decrease concomitantly. These highly divergent gradients will harm the model training and increase the risk of gradient vanishing and gradient exploding. To fully take the gradient stability into consideration, we present a new perspective to the BNNs training, regarding it as the equilibrium between the estimating error and the gradient stability. In this view, we firstly design two indicators to quantitatively demonstrate the equilibrium phenomenon. In addition, in order to balance the estimating error and the gradient stability well, we revise the original straight through estimator and propose a power function based estimator, Rectified Straight Through Estimator (ReSTE for short). Comparing to other estimators, ReSTE is rational and capable of flexibly balancing the estimating error with the gradient stability. Extensive experiments on CIFAR-10 and ImageNet datasets show that ReSTE has excellent performance and surpasses the state-of-the-art methods without any auxiliary modules or losses.