eduzhai > Applied Sciences > Engineering >

Graph Pooling with Node Proximity for Hierarchical Representation Learning

  • Save

... pages left unread,continue reading

Document pages: 10 pages

Abstract: Graph neural networks have attracted wide attentions to enable representationlearning of graph data in recent works. In complement to graph convolutionoperators, graph pooling is crucial for extracting hierarchical representationof graph data. However, most recent graph pooling methods still fail toefficiently exploit the geometry of graph data. In this paper, we propose anovel graph pooling strategy that leverages node proximity to improve thehierarchical representation learning of graph data with their multi-hoptopology. Node proximity is obtained by harmonizing the kernel representationof topology information and node features. Implicit structure-aware kernelrepresentation of topology information allows efficient graph pooling withoutexplicit eigendecomposition of the graph Laplacian. Similarities of nodesignals are adaptively evaluated with the combination of the affinetransformation and kernel trick using the Gaussian RBF function. Experimentalresults demonstrate that the proposed graph pooling strategy is able to achievestate-of-the-art performance on a collection of public graph classificationbenchmark datasets.

Please select stars to rate!


0 comments Sign in to leave a comment.

    Data loading, please wait...