Summary of A Backdoor Attack Against Link Prediction Tasks with Graph Neural Networks, by Jiazhu Dai et al.
A backdoor attack against link prediction tasks with graph neural networks
by Jiazhu Dai, Haoyu Sun
First submitted to arxiv on: 5 Jan 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 This paper proposes a new type of attack against Graph Neural Networks (GNNs) that can be used to manipulate link predictions. Specifically, it presents a backdoor attack on GNN-based link prediction tasks, demonstrating how an attacker can embed malicious patterns in the training data to mislead the model into predicting links between unlinked nodes. The attack leverages a single node as a trigger and manipulates node pairs during training to create a backdoor that activates when the trigger is linked to specific node pairs during inference. This vulnerability has significant implications for GNN-based applications, highlighting the need for improved security measures. |
| Low | GrooveSquid.com (original content) | Low Difficulty Summary GNNs are powerful tools for analyzing graph-structured data. But did you know they can be tricked into making mistakes? Researchers have found that these models can be “poisoned” to make incorrect predictions by adding special patterns called “triggers” to the training data. This is like a secret code that tells the model what answer to give, even if it’s wrong! In this paper, scientists show how to use this trick against link prediction tasks, making GNNs produce fake links between nodes that aren’t really connected. This is important because it means we need to be careful when using GNNs in real-world applications. |
Keywords
* Artificial intelligence * Gnn * Inference
