Energy-Based Models for Graph Anomaly Detection

EasyChair Preprint 15616

Abstract

Detecting anomalies in complex networks has become a pivotal focus in graph analysis. This research introduces an innovative framework that utilizes Energy-Based Models (EBMs) to efficiently identify anomalies in graph-structured data. By combining the strengths of graph neural networks (GNNs) with EBMs, the proposed method captures structural, relational, and feature-level insights to achieve highly accurate anomaly detection. Experiments on standard benchmark datasets showcase its superior performance over state-of-the-art techniques, emphasizing the approach’s robustness and effectiveness.

Keyphrases: Graph Neural Networks, energy-based model, graph anomaly detection, machine learning, utlier detection

@booklet{EasyChair:15616,
  author    = {Kin Elvard and Hoo Chang},
  title     = {Energy-Based Models for Graph Anomaly Detection},
  howpublished = {EasyChair Preprint 15616},
  year      = {EasyChair, 2024}}