PathSAGE: Spatial Graph Attention Neural Networks With Random Path Sampling
This addresses scalability and performance issues in graph neural networks for researchers and practitioners working with large, non-Euclidean datasets, though it is incremental as it builds on existing GCN and Transformer methods.
The paper tackles the problems of 'neighbor explosion' and 'over-smoothing' in deep Graph Convolutional Networks (GCNs) for non-Euclidean data by proposing PathSAGE, a single-layer model that uses random path sampling and Transformer aggregation to learn high-order topological information, achieving comparable performance to state-of-the-art models in inductive learning tasks.
Graph Convolutional Networks (GCNs) achieve great success in non-Euclidean structure data processing recently. In existing studies, deeper layers are used in CCNs to extract deeper features of Euclidean structure data. However, for non-Euclidean structure data, too deep GCNs will confront with problems like "neighbor explosion" and "over-smoothing", it also cannot be applied to large datasets. To address these problems, we propose a model called PathSAGE, which can learn high-order topological information and improve the model's performance by expanding the receptive field. The model randomly samples paths starting from the central node and aggregates them by Transformer encoder. PathSAGE has only one layer of structure to aggregate nodes which avoid those problems above. The results of evaluation shows that our model achieves comparable performance with the state-of-the-art models in inductive learning tasks.