Identity-aware Graph Neural Networks

Abstract

Message passing Graph Neural Networks (GNNs) provide a powerful modeling framework for relational data. However, the expressive power of existing GNNs is upper-bounded by the 1-Weisfeiler-Lehman (1-WL) graph isomorphism test, which means GNNs that are not able to predict node clustering coefficients and shortest path distances, and cannot differentiate between different d-regular graphs. Here we develop a class of message passing GNNs, named Identity-aware Graph Neural Networks (ID-GNNs), with greater expressive power than the 1-WL test. ID-GNN offers a minimal but powerful solution to limitations of existing GNNs. ID-GNN extends existing GNN architectures by inductively considering nodes’ identities during message passing. To embed a given node, ID-GNN first extracts the ego network centered at the node, then conducts rounds of heterogeneous message passing, where different sets of parameters are applied to the center node than to other surrounding nodes in the ego network. We further propose a simplified but faster version of ID-GNN that injects node identity information as augmented node features. Alto- gether, both versions of ID-GNN represent general extensions of message passing GNNs, where experiments show that transforming existing GNNs to ID-GNNs yields on average 40% accuracy improvement on challenging node, edge, and graph property prediction tasks; 3% accuracy improvement on node and graph classification benchmarks; and 15% ROC AUC improvement on real-world link prediction tasks. Additionally, ID-GNNs demonstrate improved or comparable performance over other task-specific graph networks.

ROLAND: Graph Learning Framework for Dynamic Graphs

“…We use NVIDIA RTX 8000 GPU in the experiments. We implement ROLAND 1 with the GraphGym library [46].…”

“…Graph Neural Networks (GNNs) are a general class of models that can perform various learning tasks on graphs. GNNs have gained tremendous success in learning from static graphs [1,9,42,45,49]. Although various GNNs have been proposed for dynamic graphs [2,22,24,30,31,34,36,39,47,50] these approaches have limitations of model design, evaluation settings and training strategies.…”

“…Although these approaches are intuitive, they do not incorporate state-of-the-art designs from static GNNs. For instance, the incorporation of skip-connections [6,10,20], batch normalization [6,11,46], edge embedding [7,43] has been beneficial for GNN message passing, but has not been explored for dynamic GNNs. To avoid re-exploring these design choices for dynamic GNNs, instead of building dynamic GNNs from scratch, a better design principle would be to start from a mature static GNN design and adapt it for dynamic graphs.…”

Improving Zorro Explanations for Sparse Observations with Dense Proxy Data

“…2. It computes a graph embedding with general graph convolutions [12], reduces the amount of nodes to one via differential pooling [13] and predicts Q-values for the feature vector of the remaining node. Eventually, the index of the largest Q-value represents the chosen action.…”

A Block-Based Adaptive Decoupling Framework for Graph Neural Networks

“…Graph Neural Networks (GNNs) have been widely used and have achieved state-of-the-art performance in many related applications, such as node classification [ 5 , 6 , 7 , 8 ], link prediction [ 9 , 10 , 11 ], and graph classification [ 12 , 13 ]. Feature propagation is a simple, efficient, and powerful GNN paradigm [ 14 , 15 ]. The main idea behind it is to obtain new node representations by stacking multiple GNN layers to aggregate the neighbor information of nodes using nonlinear transformations [ 16 ].…”