Multi-head second-order pooling for graph transformer networks

Zhe Dong, Qilong Wang, Pengfei Zhu

Abstract: Highlights•We propose a novel multi-head graph second-order pooling method for graph transformer networks.•We normalize the covariance representation with an efficient feature dropout for generality.•We fuse the first- and second-order information adaptively.•Our proposed model is superior or competitive to state-of-the-arts on six benchmarks.