Invariant Subgraph

Invariant subgraph extraction focuses on identifying the parts of a graph that are robust to variations in data distribution, crucial for improving the generalization ability of graph neural networks (GNNs). Current research emphasizes developing algorithms, often leveraging spectral methods or optimal transport theory, to identify these invariant subgraphs, with a focus on minimizing spurious information and maximizing the retention of relevant structural and semantic information. This work is significant for enhancing the robustness and reliability of GNNs across diverse applications, including knowledge graph reasoning, few-shot learning, and out-of-distribution generalization.

Papers