GaAN: Gated Attention Networks for Learning on Large and Spatiotemporal\n Graphs
Abstract
We propose a new network architecture, Gated Attention Networks (GaAN), for\nlearning on graphs. Unlike the traditional multi-head attention mechanism,\nwhich equally consumes all attention heads, GaAN uses a convolutional\nsub-network to control each attention head's importance. We demonstrate the\neffectiveness of GaAN on the inductive node classification problem. Moreover,\nwith GaAN as a building block, we construct the Graph Gated Recurrent Unit\n(GGRU) to address the traffic speed forecasting problem. Extensive experiments\non three real-world datasets show that our GaAN framework achieves\nstate-of-the-art results on both tasks.\n
Authors: Jiani Zhang, Xingjian Shi, Junyuan Xie, Hao Ma, Irwin King, Dit Yan Yeung
Published in: arXiv (Cornell University) (2018)