Summary of Arq: a Mixed-precision Quantization Framework For Accurate and Certifiably Robust Dnns, by Yuchen Yang et al.
ARQ: A Mixed-Precision Quantization Framework for Accurate and Certifiably Robust DNNs
by Yuchen Yang, Shubham Ugare, Yifan Zhao, Gagandeep Singh, Sasa Misailovic
First submitted to arxiv on: 31 Oct 2024
Categories
- Main: Machine Learning (cs.LG)
- Secondary: Cryptography and Security (cs.CR); Computer Vision and Pattern Recognition (cs.CV)
GrooveSquid.com Paper Summaries
GrooveSquid.com’s goal is to make artificial intelligence research accessible by summarizing AI papers in simpler terms. Each summary below covers the same AI paper, written at different levels of difficulty. The medium difficulty and low difficulty versions are original summaries written by GrooveSquid.com, while the high difficulty version is the paper’s original abstract. Feel free to learn from the version that suits you best!
| Summary difficulty | Written by | Summary |
|---|---|---|
| High | Paper authors | High Difficulty Summary Read the original abstract here |
| Medium | GrooveSquid.com (original content) | Medium Difficulty Summary Mixed precision quantization has become a crucial technique for optimizing the execution of deep neural networks (DNNs). Despite its importance, certified robustness, which provides provable guarantees about a model’s ability to withstand different adversarial perturbations, has rarely been addressed in quantization due to unacceptably high cost. This paper introduces ARQ, an innovative mixed-precision quantization method that not only preserves the clean accuracy of smoothed classifiers but also maintains their certified robustness using reinforcement learning and randomized smoothing. ARQ outperforms multiple state-of-the-art quantization techniques across all benchmarks and input perturbation levels, achieving the performance of original DNNs with floating-point weights while reducing instructions by 1.5%. The paper’s code is available. |
| Low | GrooveSquid.com (original content) | Low Difficulty Summary ARQ is a new way to make deep neural networks (DNNs) work better on devices that need power-saving. Right now, people have trouble making sure these networks are safe from bad things added to the pictures or words they’re looking at. This makes it hard for computers to trust what the DNN says is real or not. ARQ helps make sure these networks are both fast and safe by finding the right balance between how many instructions the computer uses and how good it is at keeping the bad stuff out. |
Keywords
* Artificial intelligence * Precision * Quantization * Reinforcement learning
