Navigating Complexity: Toward Lossless Graph Condensation via Expanding Window Matching
Authors: Yuchen Zhang, Tianle Zhang, Kai Wang, Ziyao Guo, Yuxuan Liang, Xavier Bresson, Wei Jin, Yang You
ICML 2024 | Conference PDF | Archive PDF | Plain Text | LLM Run Details
| Reproducibility Variable | Result | LLM Response |
|---|---|---|
| Research Type | Experimental | Extensive experiments verify its superiority across different datasets. Code is released at https://github.com/NUS-HPC-AI-Lab/GEOM. |
| Researcher Affiliation | Academia | 1National University of Singapore 2Hong Kong University of Science and Technology (Guangzhou) 3Emory University. |
| Pseudocode | Yes | Algorithm 1 GEOM for condensing graph. |
| Open Source Code | Yes | Code is released at https://github.com/NUS-HPC-AI-Lab/GEOM. |
| Open Datasets | Yes | We conduct experiments on three transductive datasets, i.e., Cora, Citeseer (Kipf & Welling, 2016) and Ogbn-arxiv (Hu et al., 2020), and two inductive datasets, i.e, Flickr (Zeng et al., 2019) and Reddit (Hamilton et al., 2017). These datasets are sourced from Py Torch Geometric (Fey & Lenssen, 2019), with publicly accessible splits consistently applied across all experimental setups. |
| Dataset Splits | Yes | For all five datasets, we use the public splits and setups. More details of each dataset can be found in Appendix A. We first set three condensation ratios for each dataset, consistent with the setting before (Zheng et al., 2024; Jin et al., 2021), and for Ogbn-arxiv and Reddit, we add comparisons with two additional larger condensation ratios. Dataset statistics are shown in Table 4: Training/Validation/Test columns for each dataset. |
| Hardware Specification | Yes | All experiments are conducted five times on one single Nvidia-A100-SXM4-80GB GPU. |
| Software Dependencies | No | The paper mentions sourcing datasets from 'Py Torch Geometric' and using the 'GCN model'. However, it does not specify version numbers for PyTorch Geometric, PyTorch, or any other critical software libraries or dependencies, which are necessary for full reproducibility. |
| Experiment Setup | Yes | More hyper-parameter setting details are provided in Appendix G. The specific parameter settings are outlined in Table 6, where U represent the upper bounds of the expanding window, and U denotes the upper limit of the initial expanding window, incremented by one after each condensation iteration. Notably, lr y set to 0 indicates the absence of soft labels. |