Fast Hyper-walk Gridded Convolution on Graph

The existing graph convolution methods usually suffer high computation burden, large memory requirement and intractable batch-process. In this paper, we propose a high-efficient hyper-walk gridded convolution (hyper-WGC) method to encode non-regular graph data, which overcomes all these aforementioned problems. To high-efficient capture graph topology structures, we propose random hyper-walk by taking advantages of random-walks as well as node/edge encapsulation. The random hyper-walk could greatly mitigate the problem of exponentially explosive sampling times occurred in the original random walk, while well preserving graph structures to some extent. To efficiently encode local hyper-walks around one reference node, we project hyper-walks into an order space to form image-like grid data, which more favors those conventional convolution networks. We experimentally validate the efficiency and effectiveness of our proposed hyper-WGC, which has high-efficient computation speed, and comparable or even better performance when compared with those baseline GCNs.