SinReQ: Generalized Sinusoidal Regularization for Low-Bitwidth Deep Quantized Training

TL;DR

This paper introduces SinReQ, a sinusoidal regularization technique that improves low-bitwidth deep neural network training by guiding weights toward quantization-friendly values, significantly reducing accuracy loss across various models and algorithms.

Contribution

SinReQ is a novel, generalizable regularization method that enhances multiple low-bitwidth quantized training algorithms without invasive modifications.

Findings

SinReQ improves accuracy of quantized models by up to 2.8%.

It reduces the accuracy gap to full-precision models by over 27%.

Effective across different models and bitwidth configurations.

Abstract

Deep quantization of neural networks (below eight bits) offers significant promise in reducing their compute and storage cost. Albeit alluring, without special techniques for training and optimization, deep quantization results in significant accuracy loss. To further mitigate this loss, we propose a novel sinusoidal regularization, called SinReQ1, for deep quantized training. SinReQ adds a periodic term to the original objective function of the underlying training algorithm. SinReQ exploits the periodicity, differentiability, and the desired convexity profile in sinusoidal functions to automatically propel weights towards values that are inherently closer to quantization levels. Since, this technique does not require invasive changes to the training procedure, SinReQ can harmoniously enhance quantized training algorithms. SinReQ offers generality and flexibility as it is not limited to a certain bitwidth or a uniform assignment of bitwidths across layers. We carry out experimentation using the AlexNet, CIFAR-10, ResNet-18, ResNet-20, SVHN, and VGG-11 DNNs with three to five bits for quantization and show the versatility of SinReQ in enhancing multiple quantized training algorithms, DoReFa [32] and WRPN [24]. Averaging across all the bit configurations shows that SinReQ closes the accuracy gap between these two techniques and the full-precision runs by 32.4% and 27.5%, respectively. That is improving the absolute accuracy of DoReFa and WRPN by 2.8% and 2.1%, respectively.