LMC: Fast Training of GNNs via Subgraph Sampling with Provable Convergence

Authors: Zhihao Shi, Xize Liang, Jie Wang

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

Reproducibility Variable Result LLM Response
Research Type Experimental Experiments on large-scale benchmark tasks demonstrate that LMC significantly outperforms state-of-the-art subgraph-wise sampling methods in terms of efficiency.
Researcher Affiliation Academia Zhihao Shi 1 , Xize Liang 1, Jie Wang 1, 1 University of Science and Technology of China
Pseudocode Yes Algorithm 1 Local Message Compensation
Open Source Code Yes The code of LMC is available on Git Hub at https://github.com/MIRALab-USTC/GNN-LMC.
Open Datasets Yes Therefore, we evaluate LMC on four large datasets, PPI, REDDIT , FLICKR(Hamilton et al., 2017), and Ogbn-arxiv (Hu et al., 2020).
Dataset Splits Yes We use the data splitting strategies following previous works (Fey et al., 2021; Gu et al., 2020). [...] Although GAS finally resembles full-batch performance in Table 1 by selecting the best performance on the valid data, it may fail to resemble under small batch sizes
Hardware Specification Yes We run all experiments on a single Ge Force RTX 2080 Ti (11 GB).
Software Dependencies No The paper mentions 'Pytorch (Paszke et al., 2019) and Py Torch Geometric (Fey & Lenssen, 2019)' but does not provide explicit version numbers for these software dependencies.
Experiment Setup Yes To ensure a fair comparison, we follow the data splits, training pipeline, and most hyperparameters in (Fey et al., 2021) except for the additional hyperparameters in LMC such as βi. We use the grid search to find the best βi (see Appendix A.4 for more details).