Summary of Agsoa:graph Neural Network Targeted Attack Based on Average Gradient and Structure Optimization, by Yang Chen and Bin Zhou
AGSOA:Graph Neural Network Targeted Attack Based on Average Gradient and Structure Optimization
by Yang Chen, Bin Zhou
First submitted to arxiv on: 19 Jun 2024
Categories
- Main: Machine Learning (cs.LG)
- Secondary: Artificial Intelligence (cs.AI); 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 AGSOA attack on Graph Neural Networks (GNNs) addresses the limitations of current gradient-based attacks. These attacks are prone to falling into local optima, leading to underperformance, and often prioritize effectiveness over invisibility, making them easily exposed. To overcome these issues, AGSOA combines average gradient calculation and structure optimization modules. The former stabilizes the attack direction by averaging gradient information across moments, while the latter adjusts graph structure to improve attack invisibility and transferability. Experimental results on three datasets demonstrate that AGSOA outperforms state-of-the-art models, achieving a 2%-8% increase in misclassification rate. |
| Low | GrooveSquid.com (original content) | Low Difficulty Summary The paper proposes an attack method called AGSOA to target Graph Neural Networks (GNNs). GNNs are vulnerable to attacks that add small perturbations to the graph. The current gradient-based attacks have limitations, such as easily falling into local optima and being easy to detect. AGSOA tries to solve these problems by using both average gradient calculation and structure optimization modules. It calculates the average of the gradients over all moments to guide the attack and adjusts the graph structure to make the attack harder to detect. The paper shows that AGSOA works better than other models on three datasets. |
Keywords
» Artificial intelligence » Optimization » Transferability
