Saved in:
Bibliographic Details
Main Authors: Yang, Yuchen, Zhao, Yifan, Ugare, Shubham, Singh, Gagandeep, Misailovic, Sasa
Format: Preprint
Published: 2024
Subjects:
Online Access:https://arxiv.org/abs/2410.24214
Tags: Add Tag
No Tags, Be the first to tag this record!
_version_ 1866917440984711168
author Yang, Yuchen
Zhao, Yifan
Ugare, Shubham
Singh, Gagandeep
Misailovic, Sasa
author_facet Yang, Yuchen
Zhao, Yifan
Ugare, Shubham
Singh, Gagandeep
Misailovic, Sasa
contents Mixed precision quantization has become an important technique for optimizing the execution of deep neural networks (DNNs). Certified robustness, which provides provable guarantees about a model's ability to withstand different adversarial perturbations, has rarely been addressed in quantization due to the unacceptably high cost of certifying robustness. This paper introduces ARQ, an innovative mixed-precision quantization method that not only preserves the clean accuracy of the smoothed classifiers, but also maintains their certified robustness. ARQ uses reinforcement learning to find accurate and robust DNN quantization, while efficiently leveraging randomized smoothing, a popular class of statistical DNN verification algorithms. ARQ consistently performs better than multiple state-of-the-art quantization techniques across all the benchmarks and the input perturbation levels. The performance of ARQ quantized networks reaches that of the original DNN with floating-point weights, while using only 1.5% instructions and the highest certified radius. ARQ's code is available at https://github.com/uiuc-arc/ARQ.
format Preprint
id arxiv_https___arxiv_org_abs_2410_24214
institution arXiv
publishDate 2024
record_format arxiv
spellingShingle ARQ: A Mixed-Precision Quantization Framework for Accurate and Certifiably Robust DNNs
Yang, Yuchen
Zhao, Yifan
Ugare, Shubham
Singh, Gagandeep
Misailovic, Sasa
Machine Learning
Cryptography and Security
Computer Vision and Pattern Recognition
Mixed precision quantization has become an important technique for optimizing the execution of deep neural networks (DNNs). Certified robustness, which provides provable guarantees about a model's ability to withstand different adversarial perturbations, has rarely been addressed in quantization due to the unacceptably high cost of certifying robustness. This paper introduces ARQ, an innovative mixed-precision quantization method that not only preserves the clean accuracy of the smoothed classifiers, but also maintains their certified robustness. ARQ uses reinforcement learning to find accurate and robust DNN quantization, while efficiently leveraging randomized smoothing, a popular class of statistical DNN verification algorithms. ARQ consistently performs better than multiple state-of-the-art quantization techniques across all the benchmarks and the input perturbation levels. The performance of ARQ quantized networks reaches that of the original DNN with floating-point weights, while using only 1.5% instructions and the highest certified radius. ARQ's code is available at https://github.com/uiuc-arc/ARQ.
title ARQ: A Mixed-Precision Quantization Framework for Accurate and Certifiably Robust DNNs
topic Machine Learning
Cryptography and Security
Computer Vision and Pattern Recognition
url https://arxiv.org/abs/2410.24214