Adaptive multi-scale Graph Neural Architecture Search framework

Lintao Yang, Pietro Liò, Xu Shen, Yuyang Zhang, Chengbin Peng

Abstract: Highlights•We propose a paradigm to search for message-passing and feature fusion strategies.•The proposed message-passing scheme allows a much larger search space.•Our approach can overcome the limit of the low-pass filtering in traditional GNNs.•Our approach outperforms traditional GNAS and human-designed GNNs on different tasks.