SQS: Bayesian DNN Compression through Sparse Quantized Sub-distributions

Fuente: arXiv
Saved in:
Bibliographic Details
Main Authors: Wang, Ziyi, Jiang, Nan, Lin, Guang, Song, Qifan
Format: Preprint
Published: 2025
Subjects:
Online Access:
Tags: Add Tag
No Tags, Be the first to tag this record!
_version_ 1866917001738321920
author Wang, Ziyi
Jiang, Nan
Lin, Guang
Song, Qifan
author_facet Wang, Ziyi
Jiang, Nan
Lin, Guang
Song, Qifan
contents Compressing large-scale neural networks is essential for deploying models on resource-constrained devices. Most existing methods adopt weight pruning or low-bit quantization individually, often resulting in suboptimal compression rates to preserve acceptable performance drops. We introduce a unified framework for simultaneous pruning and low-bit quantization via Bayesian variational learning (SQS), which achieves higher compression rates than prior baselines while maintaining comparable performance. The key idea is to employ a spike-and-slab prior to inducing sparsity and model quantized weights using Gaussian Mixture Models (GMMs) to enable low-bit precision. In theory, we provide the consistent result of our proposed variational approach to a sparse and quantized deep neural network. Extensive experiments on compressing ResNet, BERT-base, Llama3, and Qwen2.5 models show that our method achieves higher compression rates than a line of existing methods with comparable performance drops.
format Preprint
id arxiv_https___arxiv_org_abs_2510_08999
institution arXiv
publishDate 2025
record_format arxiv
spellingShingle SQS: Bayesian DNN Compression through Sparse Quantized Sub-distributions
Wang, Ziyi
Jiang, Nan
Lin, Guang
Song, Qifan
Machine Learning
Artificial Intelligence
Compressing large-scale neural networks is essential for deploying models on resource-constrained devices. Most existing methods adopt weight pruning or low-bit quantization individually, often resulting in suboptimal compression rates to preserve acceptable performance drops. We introduce a unified framework for simultaneous pruning and low-bit quantization via Bayesian variational learning (SQS), which achieves higher compression rates than prior baselines while maintaining comparable performance. The key idea is to employ a spike-and-slab prior to inducing sparsity and model quantized weights using Gaussian Mixture Models (GMMs) to enable low-bit precision. In theory, we provide the consistent result of our proposed variational approach to a sparse and quantized deep neural network. Extensive experiments on compressing ResNet, BERT-base, Llama3, and Qwen2.5 models show that our method achieves higher compression rates than a line of existing methods with comparable performance drops.
title SQS: Bayesian DNN Compression through Sparse Quantized Sub-distributions
topic Machine Learning
Artificial Intelligence
url https://arxiv.org/abs/2510.08999