Summary of Advancing Generalized Transfer Attack with Initialization Derived Bilevel Optimization and Dynamic Sequence Truncation, by Yaohua Liu et al.
Advancing Generalized Transfer Attack with Initialization Derived Bilevel Optimization and Dynamic Sequence Truncation
by Yaohua Liu, Jiaxin Gao, Xuan Liu, Xianghao Jiao, Xin Fan, Risheng Liu
First submitted to arxiv on: 4 Jun 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 The proposed BilLevel Transfer Attack (BETAK) framework addresses the limitations of existing transfer attacks by establishing a bilevel optimization paradigm that reformulates the nested constraint relationship between the upper-level pseudo-victim attacker and the lower-level surrogate attacker. BETAK introduces the Hyper Gradient Response (HGR) estimation as an effective feedback for transferability over pseudo-victim attackers, and proposes the Dynamic Sequence Truncation (DST) technique to dynamically adjust the back-propagation path for HGR and reduce computational overhead. The framework demonstrates substantial improvement in attack success rates against different victims and defense methods in targeted and untargeted attack scenarios. |
| Low | GrooveSquid.com (original content) | Low Difficulty Summary BETAK is a new way to create attacks that can work on different computers or models. Right now, making these attacks works well only when the attacker knows what the defender’s model looks like. The problem is that this makes it hard for the attacker to understand how their attack will work in real-life situations where they don’t know the defender’s model. BETAK solves this problem by using a special kind of optimization to make the attack work better on different models. This means that the attack can be more successful and harder to defend against. |
Keywords
» Artificial intelligence » Optimization » Transferability
