Reviews: Structure-Aware Convolutional Neural Networks

The paper proposes a structure-aware convolution structure that can naturally aggregate local inputs from diverse topological structures without predefining data structures. The learnable filters in the convolution operation are well designed in a simple way. The methods shows better quantitative results on eleven datasets when comparing to baselines. The convolution operator is simply unified on diverse topological structures. However, in equation (6), a same x_i represents a single value (with multi-channel) at i-th vertex only. Can the authors briefly compares the proposed method with [17]?