Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

GRAM: An Interpretable Approach for Graph Anomaly Detection using Gradient Attention Maps

Yifei Yang, Peng Wang, Xiaofan He, Dongmian Zou

TL;DR

This paper proposes a novel approach to graph anomaly detection that leverages the power of interpretability to enhance performance and extracts an attention map derived from gradients of graph neural networks, which serves as a basis for scoring anomalies.

Abstract

Detecting unusual patterns in graph data is a crucial task in data mining. However, existing methods face challenges in consistently achieving satisfactory performance and often lack interpretability, which hinders our understanding of anomaly detection decisions. In this paper, we propose a novel approach to graph anomaly detection that leverages the power of interpretability to enhance performance. Specifically, our method extracts an attention map derived from gradients of graph neural networks, which serves as a basis for scoring anomalies. Notably, our approach is flexible and can be used in various anomaly detection settings. In addition, we conduct theoretical analysis using synthetic data to validate our method and gain insights into its decision-making process. To demonstrate the effectiveness of our method, we extensively evaluate our approach against state-of-the-art graph anomaly detection techniques on real-world graph classification and wireless network datasets. The results consistently demonstrate the superior performance of our method compared to the baselines.

GRAM: An Interpretable Approach for Graph Anomaly Detection using Gradient Attention Maps

TL;DR

This paper proposes a novel approach to graph anomaly detection that leverages the power of interpretability to enhance performance and extracts an attention map derived from gradients of graph neural networks, which serves as a basis for scoring anomalies.

Abstract

Detecting unusual patterns in graph data is a crucial task in data mining. However, existing methods face challenges in consistently achieving satisfactory performance and often lack interpretability, which hinders our understanding of anomaly detection decisions. In this paper, we propose a novel approach to graph anomaly detection that leverages the power of interpretability to enhance performance. Specifically, our method extracts an attention map derived from gradients of graph neural networks, which serves as a basis for scoring anomalies. Notably, our approach is flexible and can be used in various anomaly detection settings. In addition, we conduct theoretical analysis using synthetic data to validate our method and gain insights into its decision-making process. To demonstrate the effectiveness of our method, we extensively evaluate our approach against state-of-the-art graph anomaly detection techniques on real-world graph classification and wireless network datasets. The results consistently demonstrate the superior performance of our method compared to the baselines.
Paper Structure (19 sections, 27 equations, 5 figures, 9 tables)

This paper contains 19 sections, 27 equations, 5 figures, 9 tables.

Table of Contents

  1. Introduction
  2. Related Works
  3. Proposed Method
  4. GRAM for anomalous graphs
  5. GRAM for anomalous nodes
  6. Analysis on a Simple Scenario
  7. Experiments
  8. GRAM for Graph-Level Anomaly Detection
  9. Datasets
  10. Baseline methods
  11. Experimental settings
  12. Results
  13. Interpretative performance
  14. GRAM for Node-Level Anomaly Detection
  15. Datasets
  16. ...and 4 more sections

Figures (5)

  • Figure 1: Illustration of graph anomaly detection (GAD) tasks. (a) A graph is anomalous within a given dataset of graphs. (b) A node is anomalous within a given graph, where the graph data may comprise multiple graphs.
  • Figure 2: The framework of the GRAM method based on the VGAE model. (a) Training phase: The VGAE model is trained according to the reconstruction error. (b) Testing phase: The encoder is employed to compute the anomaly score.
  • Figure 3: The framework of the GRAM method based on the regression model. Training phase: the regression model is trained using the MSE loss. Testing phase: the trained model is employed to produce the anomaly score.
  • Figure 4: Examples of a binary tree graph and a double ring graph.
  • Figure 5: Visualization of anomaly scores for sampled SYN data. These scores are mapped to the grayscale intensities of the graph nodes for clarity.