Advancing Graph Anomaly Detection with Energy-Based Models: a Comprehensive Framework

EasyChair Preprint 15485

Abstract

Graph anomaly detection has emerged as a critical area in understanding complex networks. This study proposes a novel framework leveraging Energy-Based Models (EBMs) to detect anomalies in graph-structured data efficiently. By integrating graph neural networks (GNNs) with EBMs, we aim to exploit structural, relational, and feature-level information to identify outliers with high accuracy. Experimental results on benchmark datasets demonstrate superior performance compared to state-of-the-art methods, highlighting the robustness of our approach.

Keyphrases: Graph Neural Networks, energy-based models, graph anomaly detection, machine learning, outlier detection

@booklet{EasyChair:15485,
  author    = {Reza Mehri and Amin Bagheri and Behdad Jafari and Mahnaz Amini and Mobina Bagheri and Mehmmet Amin},
  title     = {Advancing Graph Anomaly Detection with Energy-Based Models: a Comprehensive Framework},
  howpublished = {EasyChair Preprint 15485},
  year      = {EasyChair, 2024}}