Using Random Noise Equivariantly to Boost Graph Neural Networks Universally
This provides a general method to boost expressivity in GNNs for various graph tasks, addressing a bottleneck in noise utilization.
The paper tackled the problem of random noise degrading performance in Graph Neural Networks (GNNs) by proposing a theoretical framework on sample complexity and introducing the Equivariant Noise GNN (ENGNN) architecture, which significantly enhanced performance across node-level, link-level, subgraph, and graph-level tasks and achieved comparable results to task-specific models.
Recent advances in Graph Neural Networks (GNNs) have explored the potential of random noise as an input feature to enhance expressivity across diverse tasks. However, naively incorporating noise can degrade performance, while architectures tailored to exploit noise for specific tasks excel yet lack broad applicability. This paper tackles these issues by laying down a theoretical framework that elucidates the increased sample complexity when introducing random noise into GNNs without careful design. We further propose Equivariant Noise GNN (ENGNN), a novel architecture that harnesses the symmetrical properties of noise to mitigate sample complexity and bolster generalization. Our experiments demonstrate that using noise equivariantly significantly enhances performance on node-level, link-level, subgraph, and graph-level tasks and achieves comparable performance to models designed for specific tasks, thereby offering a general method to boost expressivity across various graph tasks.