EiG-Search: Generating Edge-Induced Subgraphs for GNN Explanation in Linear Time

Authors: Shengyao Lu, Bang Liu, Keith G Mills, Jiao He, Di Niu

ICML 2024 | Conference PDF | Archive PDF | Plain Text | LLM Run Details

Reproducibility Variable Result LLM Response
Research Type Experimental We conduct extensive experiments on a total of seven datasets, demonstrating its superior performance and efficiency both quantitatively and qualitatively over the leading baselines.
Researcher Affiliation Collaboration 1Department of Electrical and Computer Engineering, University of Alberta 2DIRO, Universit e de Montr eal & Mila, Canada CIFAR AI Chair 3Kirin AI Algorithm & Solution, Huawei.
Pseudocode Yes Algorithm 1 Linear-Complexity Search for Subgraph
Open Source Code Yes Our code is available at: https://github.com/sluxsr/Ei G-Search.
Open Datasets Yes We conduct experiments both on the synthetic dataset BA-2Motifs (Luo et al., 2020), and the real-world datasets MUTAG (Debnath et al., 1991), Mutagenicity (Kazius et al., 2005), NCI1 (Wale & Karypis, 2006).
Dataset Splits Yes The GNNs are trained with the following data splits: training set (80%), validation set (10%), testing set (10%).
Hardware Specification Yes All the experiments are conducted on Intel Core i7-10700 Processor and NVIDIA Ge Force RTX 3090 Graphics Card.
Software Dependencies No The paper mentions using GNN models like GCN and GIN, but does not specify any software libraries (e.g., PyTorch, TensorFlow) or their exact version numbers used in the experiments.
Experiment Setup Yes All the GNNs contain 3 message-passing layers and a 2-layer classifier, the hidden dimension is 32 for BA-2Motifs, BA-Shapes, and 64 for BA-Community, Tree-grid, MUTAG, Mutagenicity and NCI1.