Understanding Non-linearity in Graph Neural Networks from the Bayesian-Inference Perspective

Authors: Rongzhe Wei, Haoteng YIN, Junteng Jia, Austin R. Benson, Pan Li

NeurIPS 2022 | Conference PDF | Archive PDF | Plain Text | LLM Run Details

Reproducibility Variable Result LLM Response
Research Type Experimental Extensive evaluation on both synthetic and real datasets demonstrates our theory. Specifically, the node mis-classification errors of three citation networks with different levels of attributed information (Gaussian attributes) are shown in Fig. 1, which precisely matches the above conclusions.
Researcher Affiliation Academia Rongzhe Wei1, Haoteng Yin1, Junteng Jia2, Austin R. Benson2, Pan Li1 1 Department of Computer Science, Purdue University 2 Department of Computer Science, Cornell University
Pseudocode No The paper does not contain any section explicitly labeled “Pseudocode” or “Algorithm”, nor are there structured code-like blocks.
Open Source Code Yes Our code is available at https://github. com/Graph-COM/Bayesian_inference_based_GNN.git.
Open Datasets Yes This experiments compare non-linear models and linear models under Gaussian and Laplacian attributes on three benchmark citation networks Pub Med, Cora, and Cite Seer [92].
Dataset Splits No One graph is used for training and the other one for testing.
Hardware Specification No The provided text does not explicitly describe the hardware used for running the experiments. It only mentions a reference to Appendix G for compute resources, which is not available.
Software Dependencies No The provided text does not explicitly list any software dependencies with specific version numbers.
Experiment Setup Yes The model is trained with Adam optimizer (learning rate = 1e-2, weight decay = 5e-4).