Summary of Graph Anomaly Detection with Noisy Labels by Reinforcement Learning, By Zhu Wang et al.
Graph Anomaly Detection with Noisy Labels by Reinforcement Learning
by Zhu Wang, Shuang Zhou, Junnan Dong, Chang Yang, Xiao Huang, Shengjie Zhao
First submitted to arxiv on: 8 Jul 2024
Categories
- Main: Machine Learning (cs.LG)
- Secondary: Artificial Intelligence (cs.AI)
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 paper proposes a novel framework for graph anomaly detection, called REinforced Graph Anomaly Detector (REGAD), which aims to improve the performance of base detectors by carefully pruning noisy edges in graphs. Existing methods rely heavily on high-quality annotation, but this is challenging to obtain in real-world scenarios, leading to degraded performance due to noisy labels. REGAD addresses this issue by maximizing the performance improvement of a base detector by cutting noisy edges approximated through nodes with high-confidence labels. The framework consists of two components: a policy network that trains a policy to prune edges step-by-step, and a policy-in-the-loop mechanism that iteratively optimizes the policy based on feedback from the base detector. The overall performance is evaluated by cumulative rewards. Extensive experiments are conducted on three datasets under different anomaly ratios, showing superior performance of REGAD. |
| Low | GrooveSquid.com (original content) | Low Difficulty Summary The paper creates a new way to find strange things in graphs, like suspicious connections between people or accounts. Current methods rely too much on humans labeling the data, which can be hard and even make mistakes. This new method, called REGAD, tries to fix this problem by carefully cutting out noisy connections that might be fake or misleading. The approach uses a special type of artificial intelligence (AI) called a policy network that learns how to prune these edges step-by-step. The AI also gets feedback from another detector and adjusts its strategy based on that feedback. This new method seems to work better than current methods in real-world scenarios, according to experiments with different amounts of suspicious connections. |
Keywords
* Artificial intelligence * Anomaly detection * Pruning
