Summary of Cross-context Backdoor Attacks Against Graph Prompt Learning, by Xiaoting Lyu et al.
Cross-Context Backdoor Attacks against Graph Prompt Learning
by Xiaoting Lyu, Yufei Han, Wei Wang, Hangwei Qian, Ivor Tsang, Xiangliang Zhang
First submitted to arxiv on: 28 May 2024
Categories
- Main: Machine Learning (cs.LG)
- Secondary: None
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 Graph Prompt Learning (GPL) framework has shown promise in bridging the gap between pretraining and downstream graph learning applications. However, this paper reveals a significant vulnerability to backdoor attacks, which can be embedded during the pretraining phase. The proposed attack, CrossBA, manipulates trigger graphs and prompt transformations to transfer the backdoor threat from pretrained encoders to downstream applications. Through extensive experiments across various GPL methods, scenarios, and datasets, the study demonstrates the effectiveness of CrossBA in achieving high attack success rates while preserving application functionality. This highlights concerns about the trustworthiness of GPL techniques and underscores the need for robust countermeasures. |
| Low | GrooveSquid.com (original content) | Low Difficulty Summary In simple terms, this paper is about a new kind of threat to a type of artificial intelligence (AI) called Graph Prompt Learning. This AI helps machines understand and work with complex data structures like social networks or molecular bonds. The problem is that someone could secretly add bad information into the training data, making the AI do things it shouldn’t. The researchers created a way to test this kind of attack and found that it’s surprisingly effective. They’re warning others about this risk so we can build better ways to keep our AI systems safe. |
Keywords
» Artificial intelligence » Pretraining » Prompt
