Summary of Differentiable Search Of Accurate and Robust Architectures, by Yuwei Ou et al.
Differentiable Search of Accurate and Robust Architectures
by Yuwei Ou, Xiangning Xie, Shangce Gao, Yanan Sun, Kay Chen Tan, Jiancheng Lv
First submitted to arxiv on: 28 Dec 2022
Categories
- Main: Machine Learning (cs.LG)
- Secondary: Artificial Intelligence (cs.AI); Cryptography and Security (cs.CR)
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 The proposed DSARA method automatically searches for neural architectures that are both accurate and robust after adversarial training. This approach addresses the limitation of traditional methods by allowing the search space to be designed specifically for adversarial training. The algorithm uses a two-stage strategy, first optimizing architecture parameters to minimize the adversarial loss and then refining these parameters to balance natural and adversarial losses. Experimental results show that DSARA outperforms existing methods in terms of both accuracy and robustness. |
| Low | GrooveSquid.com (original content) | Low Difficulty Summary DSARA is a new way to find the best neural networks for protecting against bad data. Normally, this process relies on trial and error or requires expert knowledge. But DSARA uses special architecture designs and two stages of searching to find networks that are both good at recognizing normal data and resistant to attacks. The results show that DSARA finds better networks than previous methods. |
