Summary of Exploring Adversarial Robustness Of Deep State Space Models, by Biqing Qi et al.
Exploring Adversarial Robustness of Deep State Space Models
by Biqing Qi, Yang Luo, Junqi Gao, Pengfei Li, Kai Tian, Zhiyuan Ma, Bowen Zhou
First submitted to arxiv on: 8 Jun 2024
Categories
- Main: Machine Learning (cs.LG)
- Secondary: Artificial Intelligence (cs.AI)
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 paper investigates the Adversarial Robustness (AR) of Deep State Space Models (SSMs) using Adversarial Training (AT). While SSMs have shown promise in various task scenarios, their AR performance remains unclear. The authors evaluate existing structural variants of SSMs with AT and find that incorporating Attention yields a better trade-off between robustness and generalization. However, this integration also leads to Robust Overfitting (RO) issues. To address these phenomena, the authors analyze the output error of SSMs under Adversarial Perturbations (AP). They propose an Adaptive Scaling (AdS) mechanism that improves AT performance without introducing RO. The authors’ findings are supported by empirical and theoretical analyses. |
| Low | GrooveSquid.com (original content) | Low Difficulty Summary This paper looks at how well a type of AI model called Deep State Space Models can withstand attacks when used in real-world situations. These models have been shown to be effective, but they need to be made more secure. The authors tested different versions of the model and found that adding something called Attention helps make it more robust. However, this also makes it more prone to overfitting, which is a problem. To solve these issues, the authors analyzed how well the model performs when attacked and proposed a new way to improve its security without making it overfit. |
Keywords
» Artificial intelligence » Attention » Generalization » Overfitting
