一句话：提出掩码向量量化 MVQ，通过 N:M 剪枝与掩码 k-means 保护重要权重，并用稀疏脉动阵列加速 DNN 压缩。
传统向量量化（VQ）硬件友好，可降低 DNN 存储开销与权重加载位宽，但未保留重要权重，导致显著精度损失。论文提出 MVQ：算法上先用 N:M 剪枝去除不重要权重，再用掩码 k-means 仅计算未剪枝权重与码字的距离并更新码字，以有限码字更好逼近重要权重；架构上在增强权重固定（EWS）CNN 加速器上实现 VQ，并设计稀疏脉动阵列。图像分类、目标检测和分割实验表明，MVQ 在可比压缩率下优于传统 VQ 且降低 FLOPs。ASIC 评估显示，相比基础 EWS 加速器能效提升 2.3×、脉动阵列面积减少 55%，相比此前稀疏加速器能效提升 1.73×。
Vector quantization(VQ) is a hardware-friendly DNN compression method that can reduce the storage cost and weight-loading datawidth of hardware accelerators. However, conventional VQ techniques lead to significant accuracy loss because the important weights are not well preserved. To tackle this problem, a novel approach called MVQ is proposed, which aims at better approximating important weights with a limited number of codewords. At the algorithm level, our approach removes the less important weights through N:M pruning and then minimizes the vector clustering error between the remaining weights and codewords by the masked k-means algorithm. Only distances between the unpruned weights and the codewords are computed, which are then used to update the codewords. At the architecture level, our accelerator implements vector quantization on an EWS (Enhanced weight stationary) CNN accelerator and proposes a sparse systolic array design to maximize the benefits brought by masked vector quantization. Our algorithm is validated on various models for image classification, object detection, and segmentation tasks. Experimental results demonstrate that MVQ not only outperforms conventional vector quantization methods at comparable compression ratios but also reduces FLOPs. Under ASIC evaluation, our MVQ accelerator boosts energy efficiency by 2.3× and reduces the size of the systolic array by 55% when compared with the base EWS accelerator. Compared to the previous sparse accelerators, MVQ achieves 1.73× higher energy-efficiency.
DOI 原文 · arXiv · PDF（开放获取） · DBLP
@inproceedings{LiWDWYWSH25,
title = {MVQ: Towards Efficient DNN Compression and Acceleration with Masked Vector Quantization},
author = {Shuaiting Li and Chengxuan Wang and Juncan Deng and Zeyu Wang and Zewen Ye and Zongsheng Wang and Haibin Shen and Kejie Huang},
booktitle = {ASPLOS},
year = {2025},
doi = {10.1145/3669940.3707268}
}