Graph Diffusion Counterfactual Explanation
This work addresses the need for interpretability in graph-based machine learning models, such as those used in molecular graphs or social networks, by providing a method to explain predictions, which is an incremental advancement in the underexplored graph domain.
The paper tackles the problem of generating counterfactual explanations for graph-structured data, which is challenging due to graphs being discrete and non-Euclidean, by introducing a novel framework that combines discrete diffusion models and classifier-free guidance, resulting in reliable generation of in-distribution and minimally structurally different counterfactuals for both discrete and continuous targets.
Machine learning models that operate on graph-structured data, such as molecular graphs or social networks, often make accurate predictions but offer little insight into why certain predictions are made. Counterfactual explanations address this challenge by seeking the closest alternative scenario where the model's prediction would change. Although counterfactual explanations are extensively studied in tabular data and computer vision, the graph domain remains comparatively underexplored. Constructing graph counterfactuals is intrinsically difficult because graphs are discrete and non-euclidean objects. We introduce Graph Diffusion Counterfactual Explanation, a novel framework for generating counterfactual explanations on graph data, combining discrete diffusion models and classifier-free guidance. We empirically demonstrate that our method reliably generates in-distribution as well as minimally structurally different counterfactuals for both discrete classification targets and continuous properties.