A Research On The Structural Inference Ability Of Graph Convolutional Neural Networks

Graph-structured data are ubiquitous in the natural and social sciences.In recent years,there has been a proliferation of research on graph deep learning,including deep graph embedding techniques,the generalization of convolutional neural networks to graph-structured data,and neural information transfer methods inspired by information propagation.These advances in graph deep learning have yielded state-of-the-art results in many fields,including chemical synthesis,computer vision,recommendation systems,and social network analysis.In this paper,we analyze the ability of graph neural networks to capture higher-level graph information in terms of their powerful information extraction capabilities,and the application of graph neural networks to timing optimization.The main work of this paper is as follows:(1)For graph classification tasks,existing work on graph classification mainly focuses on two aspects of graph similarity: physical structure and practical property.In this paper,we consider the problem of graph classification from a new perspective,namely structural properties.Graph similarity is defined based on structural properties such as maximum clique,minimum vertex coverage,and minimum dominating set of graphs.To capture these structural features,we design an adaptive motif to mine the higher-order connectivity information among nodes.Furthermore,to obtain the unique down-sampling in graph pooling stage,we propose a decorrelation pooling approach.Our extensive experiments on several artificially generated datasets show that our proposed model can effectively classify graphs with similar structural property.It is also experimentally compared with the baseline approach to demonstrate the effectiveness of our adaptive motif graph convolutional neural networks.(2)Since graph neural networks(GNNs)are often designed as an end-to-end learning framework,it is a fundamental issue to study the impact of graph topological structure and node features on the learning performance of graph neural networks.This paper first presents several experiments to evaluate the influence of graph topological structure and node features on the learning performance of GNNs.On this basis,four optional methods for graph-structured data augmentation are proposed from the perspective of graph topological structure and feature matrix.Meanwhile,in some high-dimensional problems,we cannot directly rely on the supervised information,so this paper leverage consistent regular terms to constrain unsupervised information.Experiments on various graph classification datasets demonstrate the effectiveness of our proposed methods.(3)Timing optimization is a key flow in electronic design automation(EDA)tools,with the goal of ensuring that chip designs are functionally correct and perform to design requirements.Timing constraints typically include three aspects: timing design rule constraints,hold time constraints,and setup time constraints.Timing violations occur when these constraints are not met,and timing optimization is a tool for optimizing timing violations.For timing graphs,which naturally have the characteristics and properties of graphs,this paper introduces the common ways to optimize timing violations,and explores the application prospects of graph neural networks algorithm in timing optimization.