English

Cluster-Promoting Quantization with Bit-Drop for Minimizing Network Quantization Loss

Machine Learning 2021-09-07 v1 Computer Vision and Pattern Recognition

Abstract

Network quantization, which aims to reduce the bit-lengths of the network weights and activations, has emerged for their deployments to resource-limited devices. Although recent studies have successfully discretized a full-precision network, they still incur large quantization errors after training, thus giving rise to a significant performance gap between a full-precision network and its quantized counterpart. In this work, we propose a novel quantization method for neural networks, Cluster-Promoting Quantization (CPQ) that finds the optimal quantization grids while naturally encouraging the underlying full-precision weights to gather around those quantization grids cohesively during training. This property of CPQ is thanks to our two main ingredients that enable differentiable quantization: i) the use of the categorical distribution designed by a specific probabilistic parametrization in the forward pass and ii) our proposed multi-class straight-through estimator (STE) in the backward pass. Since our second component, multi-class STE, is intrinsically biased, we additionally propose a new bit-drop technique, DropBits, that revises the standard dropout regularization to randomly drop bits instead of neurons. As a natural extension of DropBits, we further introduce the way of learning heterogeneous quantization levels to find proper bit-length for each layer by imposing an additional regularization on DropBits. We experimentally validate our method on various benchmark datasets and network architectures, and also support a new hypothesis for quantization: learning heterogeneous quantization levels outperforms the case using the same but fixed quantization levels from scratch.

Keywords

Cite

@article{arxiv.2109.02100,
  title  = {Cluster-Promoting Quantization with Bit-Drop for Minimizing Network Quantization Loss},
  author = {Jung Hyun Lee and Jihun Yun and Sung Ju Hwang and Eunho Yang},
  journal= {arXiv preprint arXiv:2109.02100},
  year   = {2021}
}

Comments

Accepted to ICCV 2021

R2 v1 2026-06-24T05:41:45.149Z