Dynamic Edge-Conditioned Filters in Convolutional Neural Networks on Graphs Article Swipe
YOU?
·
· 2017
· Open Access
·
· DOI: https://doi.org/10.1109/cvpr.2017.11
· OA: W2606202972
A number of problems can be formulated as prediction on graph-structured\ndata. In this work, we generalize the convolution operator from regular grids\nto arbitrary graphs while avoiding the spectral domain, which allows us to\nhandle graphs of varying size and connectivity. To move beyond a simple\ndiffusion, filter weights are conditioned on the specific edge labels in the\nneighborhood of a vertex. Together with the proper choice of graph coarsening,\nwe explore constructing deep neural networks for graph classification. In\nparticular, we demonstrate the generality of our formulation in point cloud\nclassification, where we set the new state of the art, and on a graph\nclassification dataset, where we outperform other deep learning approaches. The\nsource code is available at https://github.com/mys007/ecc\n
