Abstract: Through recognizing causal subgraphs, causal graph learning (CGL) has risen to be a promising approach for improving the generalizability of graph neural networks under out-of-distribution (OOD) scenarios. However, the empirical successes of CGL techniques are mostly exemplified in classification settings, while regression tasks, a more challenging setting in graph learning, are overlooked. We thus devote this work to tackling causal graph regression (CGR); to this end we reshape the processing of confounding effects in existing CGL studies, which mainly deal with classification. Specifically, we reflect on the predictive power of confounders in graph-level regression, and generalize classification-specific causal intervention techniques to regression through a lens of contrastive learning. Extensive experiments on graph OOD benchmarks validate the efficacy of our proposals for CGR. The model implementation and the code are provided on https://github.com/causal-graph/CGR.
Lay Summary: AI systems learning from complex networks (graphs) often struggle when faced with unfamiliar data. While "causal learning" helps these systems focus on true causes, making them more robust, it has primarily been applied to categorizing items rather than predicting specific numbers—a tougher challenge for AI on graphs.
Our research introduces a new method for applying causal learning to predict numbers on graphs. This approach addresses misleading information, known as "confounders," by adapting established techniques. Furthermore, we utilize "contrastive learning." This enables us to successfully extend causal methods, which were previously specific to classification (sorting items), to the task of regression (predicting numerical values).
The result is AI that can make more reliable numerical predictions on graphs, even as data changes. This is crucial for developing trustworthy AI in fields like materials science or economics, where data is inherently complex and ever-evolving.
Application-Driven Machine Learning: This submission is on Application-Driven Machine Learning.
Primary Area: General Machine Learning->Causality
Keywords: graph
Submission Number: 3487
Loading