Universal Graph Convolutional Networks

Authors: Di Jin, Zhizhi Yu, Cuiying Huo, Rui Wang, Xiao Wang, Dongxiao He, Jiawei Han

NeurIPS 2021 | Conference PDF | Archive PDF | Plain Text | LLM Run Details

Reproducibility Variable Result LLM Response
Research Type Experimental Extensive experiments demonstrate the superiority of U-GCN over state-of-the-arts. Extensive experiments on a series of benchmark datasets demonstrate the superiority of U-GCN over some state-of-the-arts.
Researcher Affiliation Academia 1College of Intelligence and Computing, Tianjin University, Tianjin, China 2School of Computer Science (National Pilot Software Engineering School), Beijing University of Posts and Telecommunications, Beijing, China 3Department of Computer Science, University of Illinois at Urbana-Champaign, Champaign, IL, USA
Pseudocode No The paper describes the proposed methods using mathematical formulations and textual descriptions, but it does not include any explicitly labeled 'Pseudocode' or 'Algorithm' blocks.
Open Source Code Yes The code and data are available at https://github.com/jindi-tju.
Open Datasets Yes We adopt eight public network datasets with edge homophily ratio α ranging from strong homophily to strong heterophily, as shown in Table 1, to evaluate the performance of different methods. We use three citation networks Cora, Cite Seer and Pub Med [19, 25], two Wikipedia networks Chameleon and Squirrel [24], and three webpage networks1 Cornell, Wisconsin and Texas.
Dataset Splits Yes For all methods, we set the dropout rate to 0.6 and use the same splits for training, validation and testing sets.
Hardware Specification No The paper does not provide specific details about the hardware specifications (e.g., CPU, GPU models, memory) used for running the experiments.
Software Dependencies No The paper mentions optimizers (Adam) and activation functions (ReLU, Sigmoid, Leaky ReLU), but it does not specify any software dependencies with version numbers (e.g., Python, PyTorch, TensorFlow versions, or library versions).
Experiment Setup Yes For all methods, we set the dropout rate to 0.6 and use the same splits for training, validation and testing sets. We run 5 times with the same partition and report the average results. We employ the Adam optimizer with the learning rate setting to 0.005 and apply early stopping with a patience of 20. In addition, we set the number of attention heads to 8, weight decay {5e 3, 5e 4}, and k {3...7} for k-nearest neighbor network.