Graph U-Nets Hongyang Gao and Shuiwang Ji Texas A&M University - PowerPoint PPT Presentation

Graph U-Nets Hongyang Gao and Shuiwang Ji Texas A&M University Graph U-Nets - Department of Computer Science & Engineering 1

IMAGE VS. GRAPH ­ Image can be treated as a special graph with well-defined locality. There is no locality information on normal graph, which makes it hard to define pooling and un-pooling operation on graph data. ­ Node classification problems can be considered as image segmentation problems. Both predict for each node or pixel. Graph U-Nets - Department of Computer Science & Engineering 2

Node classification Image segmentation U-NET ON GRAPH Conv layer ­ GCN layer Pooling layer ­ ? Un-pooling layer ­ ? https://lmb.informatik.uni-freiburg.de/people/ronneber/u-net/ Graph U-Nets - Department of Computer Science & Engineering 3

GRAPH POOLING LAYER (GPOOL) sigmoid 1 ! idx % ℓ ⨀ × $ # " % ℓ'( % # top k ! * ℓ * ℓ'( Inputs Projection Gate Outputs Top k Node Selection Graph U-Nets - Department of Computer Science & Engineering 4

GRAPH UN-POOLING LAYER (GUNPOOL) gPool gUnpool GCN gUnpool layer uses position information from gPool layer to reconstruct original graph structure. Graph U-Nets - Department of Computer Science & Engineering 5

GRAPH U-NET GCN GCN Inputs gPool gUnpool GCN GCN Network Embedding gPool gUnpool GCN Graph U-Nets - Department of Computer Science & Engineering 6

NETWORK REPRESENTATION LEARNING RESULTS Results on node classification tasks: Results on graph classification tasks: Graph U-Nets - Department of Computer Science & Engineering 7

GRAPH U-NETS Come to poster #25 for more details! Graph U-Nets - Department of Computer Science & Engineering 8