Summary of Universally Robust Graph Neural Networks by Preserving Neighbor Similarity, By Yulin Zhu et al.
Universally Robust Graph Neural Networks by Preserving Neighbor Similarity
by Yulin Zhu, Yuni Lai, Xing Ai, Kai Zhou
First submitted to arxiv on: 18 Jan 2024
Categories
- Main: Machine Learning (cs.LG)
- Secondary: Cryptography and Security (cs.CR); Social and Information Networks (cs.SI)
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 A surge in research has focused on enhancing the adversarial robustness of graph neural networks (GNNs) on homophilic graphs, where nodes with similar features are connected. However, it is unclear how GNNs perform on heterophilic graphs, where dissimilar nodes are linked. This paper bridges this gap by theoretically proving that the update of negative classification loss is negatively correlated with pairwise similarities based on powered aggregated neighbor features. The findings explain empirical observations that attackers tend to connect node pairs with similar neighbor features rather than ego features on both homophilic and heterophilic graphs. To address these vulnerabilities, a novel robust model called NSPGNN is introduced, incorporating a dual-kNN graph pipeline to supervise neighbor similarity-guided propagation. This approach utilizes low-pass and high-pass filters to smooth and discriminate node pair features along positive and negative kNN graphs, respectively. Extensive experiments on both homophilic and heterophilic graphs validate the universal robustness of NSPGNN compared to state-of-the-art methods. |
| Low | GrooveSquid.com (original content) | Low Difficulty Summary Graph neural networks are good at recognizing patterns in data that looks similar, but they can be tricked by attackers who create fake connections between very different things. This is a problem because it makes it hard for GNNs to work well on all kinds of graphs. To solve this problem, researchers have been trying to make GNNs more robust and less vulnerable to attacks. But so far, nobody has really looked at how GNNs do when the connections are between very different things. This paper fills that gap by showing that attackers can trick GNNs in the same way on both types of graphs. To fix this problem, the researchers came up with a new kind of GNN that is better at recognizing patterns and ignoring fake connections. |
Keywords
* Artificial intelligence * Classification * Gnn
