EUVIMEDEUVIMED entdecken
Open Access · CC BY 4.0

ARQ: A Mixed-Precision Quantization Framework for Accurate and Certifiably Robust DNNs

Yuchen Yang, Yifan Zhao, Shubham Ugare, Gagandeep Singh, Sasa Misailovic

Proceedings of the ACM on Software Engineering · 01.10.2026 · 3 (ISSTA) · 4002-4023

DOI: 10.1145/3832266 ↗

Kurz gefasst

Fachartikel in der Zeitschrift Proceedings of the ACM on Software Engineering, online erschienen am 01.10.2026. Verfasst von Yuchen Yang und 4 weiteren Autor:innen. Die Arbeit wurde bisher 2-mal zitiert (Crossref) und nennt 82 Literaturangaben. Der Volltext ist frei zugänglich (Open Access, CC BY 4.0). Gefördert von: National Science Foundation.

AbstractBibliografische AngabenZitierenVerwandte Studien

Abstract (Originalsprache)

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.

Förderung

  • National Science Foundation

Zitieren

APA

Yang, Y., Zhao, Y., Ugare, S., Singh, G., & Misailovic, S. (2026). ARQ: A Mixed-Precision Quantization Framework for Accurate and Certifiably Robust DNNs. Proceedings of the ACM on Software Engineering, 3(ISSTA), 4002-4023. https://doi.org/10.1145/3832266

Vancouver

Yang Y, Zhao Y, Ugare S, Singh G, Misailovic S. ARQ: A Mixed-Precision Quantization Framework for Accurate and Certifiably Robust DNNs. Proceedings of the ACM on Software Engineering. 2026;3(ISSTA):4002-4023. doi:10.1145/3832266

BibTeX
@article{Yang2026,
  author = {Yang, Yuchen and Zhao, Yifan and Ugare, Shubham and Singh, Gagandeep and Misailovic, Sasa},
  title = {ARQ: A Mixed-Precision Quantization Framework for Accurate and Certifiably Robust DNNs},
  journal = {Proceedings of the ACM on Software Engineering},
  year = {2026},
  volume = {3},
  number = {ISSTA},
  pages = {4002--4023},
  doi = {10.1145/3832266}
}
Kostenlos starten →