GraphMix: Improved Training of GNNs for Semi-Supervised Learning

Authors: Vikas Verma, Meng Qu, Kenji Kawaguchi, Alex Lamb, Yoshua Bengio, Juho Kannala, Jian Tang10024-10032

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

Reproducibility Variable Result LLM Response
Research Type Experimental We experimentally validate this analysis by applying Graph Mix to various architectures such as Graph Convolutional Networks, Graph Attention Networks and Graph-U-Net. Despite its simplicity, we demonstrate that Graph Mix can consistently improve or closely match stateof-the-art performance using even simpler architectures such as Graph Convolutional Networks, across three established graph benchmarks: Cora, Citeseer and Pubmed citation network datasets, as well as three newly proposed datasets: Cora Full, Co-author-CS and Co-author-Physics.
Researcher Affiliation Academia 1 Mila Québec Artificial Intelligence Institute, Montréal, Canada, 2 Aalto University, Finland, 3 Massachusetts Institute of Technology (MIT), USA
Pseudocode Yes A diagram illustrating Graph Mix is presented in Figure 1 and the full algorithm is presented in Appendix A.3.
Open Source Code Yes Code available at https://github.com/vikasverma1077/Graph Mix
Open Datasets Yes across three established graph benchmarks: Cora, Citeseer and Pubmed citation network datasets, as well as three newly proposed datasets: Cora Full, Co-author-CS and Co-author-Physics. We use Cora-Full dataset proposed in (Bojchevski and Günnemann 2018) and Coauthor-CS and Coauthor-Physics datasets proposed in (Shchur et al. 2018).
Dataset Splits Yes Along these lines, we created 10 random splits of the Cora, Citeseer and Pubmed with the same train/ validation/test number of samples as in the standard split.
Hardware Specification No The paper does not explicitly describe the specific hardware (e.g., GPU/CPU models) used for running the experiments.
Software Dependencies No The paper does not provide specific version numbers for software dependencies or libraries used in the experiments.
Experiment Setup No The paper states 'Refer to Appendix A.8 for implementation and hyperparameter details', deferring the specific experimental setup information to an appendix rather than providing it in the main text.