Sep 23, 2020 – 11:00 – 11:45 AM
Spectral clustering (SC) is a popular clustering technique to find strongly connected communities on a graph. It can be used in Graph Neural Networks to implement pooling operations that aggregate nodes belonging to the same cluster.
However, SC is memory and computational expensive and, since clustering results are graph-specific, pooling methods based on SC must perform a new optimization for each new sample. A possible solution is a GNN-based clustering approach which can address these limitations. Given a continuous relaxation of the normalized minCUT problem, which is SC-correlated, a GNN can be trained to compute cluster assignments which minimize this objective. This GNN-based implementation is differentiable, does not require to compute the spectral decomposition, and learns a clustering function that can be quickly evaluated on out-of-sample graphs. From the proposed clustering method, a graph pooling operator is presented, that overcomes some important limitations of state-of-the-art graph pooling techniques and achieves the best performance in several supervised and unsupervised tasks.
– Filippo Maria Bianchi, Daniele Grattarola, Cesare Alippi. “Spectral Clustering with Graph Neural Networks for Graph Pooling”.