The Intelligible and Effective Graph Neural Additive Networks

2024-06-03

Neural Information Processing Systems

AI Summary

Existing graph neural networks (GNNs) are predominantly black-box models, and post-hoc explanation methods fail to meet the stringent transparency requirements of high-stakes applications. Method: We propose the Graph Neural Additive Network (GNAN), the first framework to systematically extend the generalized additive model (GAM) paradigm to graph-structured data, yielding an end-to-end interpretable GNN architecture. GNAN employs differentiable piecewise-linear basis functions to model additive contributions of node and edge features, enabling direct, intrinsic global and local, feature-level and graph-level visual explanations—without requiring post-hoc attribution. Contribution/Results: On multiple benchmark graph learning tasks, GNAN achieves accuracy comparable to state-of-the-art black-box GNNs while providing fully human-understandable, complete explanation pathways. This substantially enhances model trustworthiness and is particularly suited for high-transparency domains such as clinical diagnosis and financial risk assessment.

Technology Category

• Machine Learning: Graph-based Machine Learning
• Computer Vision: Interpretability, Explainability, and Transparency
• Reasoning under Uncertainty: Graphical Models

Application Category

• Graph Algorithms and Modeling for the Web: Graph neural networks and deep learning approaches for Web-related graphs
• Semantics and Knowledge: Methods to enhance, augment, integrate or synergize semantic models such as knowledge graphs and LLMs
• User Modeling, Personalization and Recommendation: Explainable and interpretable methods for personalization

Abstract

Graph Neural Networks (GNNs) have emerged as the predominant approach for learning over graph-structured data. However, most GNNs operate as black-box models and require post-hoc explanations, which may not suffice in high-stakes scenarios where transparency is crucial. In this paper, we present a GNN that is interpretable by design. Our model, Graph Neural Additive Network (GNAN), is a novel extension of the interpretable class of Generalized Additive Models, and can be visualized and fully understood by humans. GNAN is designed to be fully interpretable, offering both global and local explanations at the feature and graph levels through direct visualization of the model. These visualizations describe exactly how the model uses the relationships between the target variable, the features, and the graph. We demonstrate the intelligibility of GNANs in a series of examples on different tasks and datasets. In addition, we show that the accuracy of GNAN is on par with black-box GNNs, making it suitable for critical applications where transparency is essential, alongside high accuracy.

Problem

• Addressing black-box nature of Graph Neural Networks
• Providing interpretable graph learning with direct visualization
• Ensuring model transparency without sacrificing predictive accuracy

Innovation

• Extends Generalized Additive Models for graphs
• Provides direct visualization for full interpretability
• Maintains accuracy comparable to black-box GNNs