Provable Training for Graph Contrastive Learning

Authors: Yue Yu, Xiao Wang, Mengmei Zhang, Nian Liu, Chuan Shi

NeurIPS 2023 | Conference PDF | Archive PDF | Plain Text | LLM Run Details

Reproducibility Variable Result LLM Response
Research Type Experimental Through extensive experiments on various benchmarks, POT consistently improves the existing GCL approaches, serving as a friendly plugin.
Researcher Affiliation Academia Yue Yu1, Xiao Wang2 , Mengmei Zhang1, Nian Liu1, Chuan Shi1 1Beijing University of Posts and Telecommunications, China 2 Beihang University, China
Pseudocode Yes Algorithm 1: Provable Training for GCL
Open Source Code Yes The complete implementation can be found at https://github.com/Void Haruhi/POT-GCL. We also provide an implementation based on Gamma GL [12] at https://github.com/BUPT-GAMMA/Gamma GL.
Open Datasets Yes We obtain the datasets from Py G [3]. Although the datasets are available for public use, we cannot find their licenses. The datasets can be found in the URLs below: Cora, Cite Seer, Pub Med: https://github.com/kimiyoung/planetoid/raw/master/data Blog Catalog: https://docs.google.com/uc?export=download&id=178PqGqh67RUYMMP6SoRHDoIBh8ku5FS&confirm=t Flickr: https://docs.google.com/uc?export=download&id=1tZp3EB20fAC27SYWwax66_8uGsuU62X&confirm=t Computers, Photo: https://github.com/shchur/gnn-benchmark/raw/master/data/npz/ Wiki CS: https://github.com/pmernyei/wiki-cs-dataset/raw/master/dataset
Dataset Splits Yes For datasets with a public split available [28], including Cora, Cite Seer, and Pub Med, we follow the public split; For other datasets with no public split, we generate random splits, where each of the training set and validation set contains 10% nodes of the graph and the rest 80% nodes of the graph is used for testing.
Hardware Specification Yes OS: Linux 5.4.0-131-generic CPU: Intel(R) Xeon(R) Gold 6348 CPU @ 2.60GHz GPU: Ge Force RTX 3090
Software Dependencies No The paper mentions implementing with 'Py Torch' and using 'Py G' for datasets, but does not specify version numbers for these software components.
Experiment Setup Yes Table 4: Hyperparameters: (p1 e, p2 e) Models Cora Cite Seer Pub Med Flickr Blog Catalog Computers Photo Wiki CS... Table 5: Hyperparameters: (τ, κ) Models Cora Cite Seer Pub Med Flickr Blog Catalog Computers Photo Wiki CS... Table 6: Hyperparameters: (pot_batch, num_epochs) Models Cora Cite Seer Pub Med Flickr Blog Catalog Computers Photo Wiki CS