Topology-Aware Gaussian Graph Repair for Robust Graph Neural Networks
Researchers have introduced Topology-Aware Gaussian Repair (TAGR), a novel framework designed to enhance the robustness of Graph Neural Networks (GNNs). TAGR addresses common issues in real-world graph data, such as noisy or missing edges, by constructing a sparse feature-neighborhood graph using an adaptive Gaussian kernel. This approach combines feature similarity with a topology-aware residual correction to repair the graph structure without requiring dense adjacency matrix learning. Experiments on citation networks demonstrate that TAGR significantly improves GNN performance under various graph imperfection scenarios. AI
IMPACT Enhances GNN performance on imperfect graph data, potentially improving real-world applications.