Jul 24, 2023
Graph neural networks (GNNs) have emerged as a powerful paradigm to learn from relational data mostly through applying the message passing mechanism. However, recent works have pointed out that the paradigm may lead to poor performance in cases of heterophily, over-smoothing, or information bottleneck. In this work, we focus on understanding and alleviating the effect of graph structural noise on GNN performance. To evaluate the graph structural noise in real data, we propose edge signal-to-noise ratio (ESNR), a novel metric evaluating overall edge noise level with respect to data features or labels based on random matrix theory. We have found striking concordance between the proposed ESNR metric and the GNN performance in various simulated and real data. To reduce the effect of the noise, we propose GPS (Graph Propensity Score) graph rewiring, which formulates a self-supervised learning task to estimate edge likelihood, which is further used to rewire the graph. We provide a theoretical guarantee for GPS graph rewiring and demonstrate its efficacy by comprehensive benchmarks.Graph neural networks (GNNs) have emerged as a powerful paradigm to learn from relational data mostly through applying the message passing mechanism. However, recent works have pointed out that the paradigm may lead to poor performance in cases of heterophily, over-smoothing, or information bottleneck. In this work, we focus on understanding and alleviating the effect of graph structural noise on GNN performance. To evaluate the graph structural noise in real data, we propose edge signal-to-nois…
Professional recording and live streaming, delivered globally.
Presentations on similar topic, category or speaker