Towards Efficient Verification of Quantized Neural Networks

Authors

  • Pei Huang Stanford University, Stanford, CA, USA
  • Haoze Wu Stanford University, Stanford, CA, USA
  • Yuting Yang Institute of Computing Technology, Chinese Academy of Sciences, Beijing, China
  • Ieva Daukantas IT University of Copenhagen, Copenhagen, Denmark
  • Min Wu Stanford University, Stanford, CA, USA
  • Yedi Zhang National University of Singapore, Singapore
  • Clark Barrett Stanford University, Stanford, CA, USA

DOI:

https://doi.org/10.1609/aaai.v38i19.30108

Keywords:

General

Abstract

Quantization replaces floating point arithmetic with integer arithmetic in deep neural network models, providing more efficient on-device inference with less power and memory. In this work, we propose a framework for formally verifying the properties of quantized neural networks. Our baseline technique is based on integer linear programming which guarantees both soundness and completeness. We then show how efficiency can be improved by utilizing gradient-based heuristic search methods and also bound-propagation techniques. We evaluate our approach on perception networks quantized with PyTorch. Our results show that we can verify quantized networks with better scalability and efficiency than the previous state of the art.
AAAI-24 / IAAI-24 / EAAI-24 Proceedings Cover

Downloads

Published

2024-03-24

How to Cite

Huang, P., Wu, H., Yang, Y., Daukantas, I., Wu, M., Zhang, Y., & Barrett, C. (2024). Towards Efficient Verification of Quantized Neural Networks. Proceedings of the AAAI Conference on Artificial Intelligence, 38(19), 21152-21160. https://doi.org/10.1609/aaai.v38i19.30108

Issue

Vol. 38 No. 19: AAAI-24 Special Track Safe, Robust and Responsible AI Track

Section

AAAI Technical Track on Safe, Robust and Responsible AI Track