Summary of Shortcuts Everywhere and Nowhere: Exploring Multi-trigger Backdoor Attacks, by Yige Li et al.
Shortcuts Everywhere and Nowhere: Exploring Multi-Trigger Backdoor Attacks
by Yige Li, Jiabo He, Hanxun Huang, Jun Sun, Xingjun Ma, Yu-Gang Jiang
First submitted to arxiv on: 27 Jan 2024
Categories
- Main: Machine Learning (cs.LG)
- Secondary: 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 study investigates the concept of Multi-Trigger Backdoor Attacks (MTBAs) in deep neural networks, which involve multiple adversaries using different types of triggers to poison a dataset. The researchers demonstrate that MTBAs can coexist, overwrite, or cross-activate one another, breaking the prevalent shortcut assumption underlying most existing backdoor detection/removal methods. To facilitate future research on detecting and mitigating these attacks, a multi-trigger backdoor poisoning dataset is created, along with potential defense strategies discussed. |
| Low | GrooveSquid.com (original content) | Low Difficulty Summary Backdoor attacks are a growing threat to deep neural networks. Normally, these attacks create shortcuts between specific classes. But what if multiple attackers used different triggers to poison the same data? That’s what Multi-Trigger Backdoor Attacks (MTBAs) do. The researchers explored three types of MTBAs: parallel, sequential, and hybrid. They showed that MTBAs can coexist, overwrite, or cross-activate each other. This breaks most existing backdoor detection methods. To help find solutions, the team created a dataset for future research. |
