Oscillations Make Neural Networks Robust to Quantization

TMLR Paper5571 Authors

Abstract: We challenge the prevailing view that weight oscillations observed during Quantization Aware Training (QAT) are merely undesirable side-effects and argue instead that they are an essential part of QAT. We show in a linear model with a single weight that the straight-through estimator (STE) results in an additional loss term that causes oscillations by pushing weights away from their nearest quantization level. Based on the mechanism from the analysis, we then derive a regularizer that induces oscillations in the weights of neural networks during training. Our empirical results on ResNet-18 and Tiny ViT on CIFAR-10 and Tiny-ImageNet datasets demonstrate across a range of quantization levels that training with oscillations followed by post-training quantization (PTQ) is sufficient to recover the performance of QAT in most cases. With this work we shed further light on the dynamics of QAT and contribute a novel insight into explaining the role of oscillations in QAT which until now have been considered to have a primarily negative effect on quantization.

Submission Length: Regular submission (no more than 12 pages of main content)

Assigned Action Editor: ~Tatiana_Likhomanenko1

Submission Number: 5571