Balancing performance and complexity with adaptive graph coarsening
Keywords: Graph representation learning, Graph coarsening, Performance-complexity trade-off, HARP
TL;DR: An approach to reducting graph size based on local graph properties such that the negative impact on model performance is minimized.
Abstract: We present a method for graph node classification that allows a user to precisely select the resolution at which the graph in question should be simplified and through this provides a way of choosing a suitable point in the performance-complexity trade-off. The method is based on refining a reduced graph in a targeted way following the node classification confidence for particular nodes.
Submission Number: 223
Loading