Summary of Mpat: Building Robust Deep Neural Networks Against Textual Adversarial Attacks, by Fangyuan Zhang et al.
MPAT: Building Robust Deep Neural Networks against Textual Adversarial Attacks
by Fangyuan Zhang, Huichi Zhou, Shuangjiao Li, Hongtao Wang
First submitted to arxiv on: 29 Feb 2024
Categories
- Main: Machine Learning (cs.LG)
- Secondary: Computation and Language (cs.CL); 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 malicious perturbation based adversarial training method (MPAT) aims to build robust deep neural networks for natural language processing tasks, effectively defending against textual adversarial attacks. By generating adversarial examples with malicious perturbations and using them instead of original inputs for model training, MPAT ensures the defense goal is achieved without compromising performance on the original task. Comprehensive experiments demonstrate that MPAT outperforms previous defense methods in defending against malicious attacks while maintaining or improving performance. |
| Low | GrooveSquid.com (original content) | Low Difficulty Summary This paper proposes a new way to make deep neural networks more secure against fake text attacks. The method creates “malicious” examples that are used to train the model, making it better at detecting and resisting these attacks. The results show that this approach is more effective than previous methods while still allowing the model to perform well on its original task. |
Keywords
* Artificial intelligence * Natural language processing
