1 INTRODUCTION
Originating from spectral graph analysis and fueled by the success of machine learning, graph neural networks (GNNs) have drawn a surge of interest and have been applied to various applications involving non-Euclidean graph-structured data. During the past few years, a wide range of GNN models [3], [10], [12], [38] have been proposed to solve graph-related problems. Exciting progress has been achieved by GNNs in domains such as recommendation systems [41], relation prediction [7], chemistry analysis [45], financial security [49], protein discovery [9], [36], EDA [16], [26], [27] and so on.