Notice: The reproducibility variables underlying each score are classified using an automated LLM-based pipeline, validated against a manually labeled dataset. LLM-based classification introduces uncertainty and potential bias; scores should be interpreted as estimates. Full accuracy metrics and methodology are described in Coakley et alK. L. Coakley, T. Snelleman, H. Hoos, and O. E. Gundersen, "The embrace of open science: An analysis of a decade of AI research and 56 800 conference papers," Under Review, 2026..
Towards Pre-trained Graph Condensation via Optimal Transport
Authors: Yeyu Yan, Shuai Zheng, Wenjun Hui, Xiangkai Zhu, Chen Dong, Zhenfeng Zhu, Yao Zhao, Kunlun He
NeurIPS 2025 | Venue PDF | LLM Run Details
| Reproducibility Variable | Result | LLM Response |
|---|---|---|
| Research Type | Experimental | Extensive experiments verify the superiority and versatility of Pre GC, demonstrating its task-independent nature and seamless compatibility with arbitrary GNNs. ... 5 Experiments |
| Researcher Affiliation | Academia | 1Institute of Information Science, Beijing Jiaotong University, China 2Visual Intelligence + X International Cooperation Joint Laboratory of the Ministry of Education, China 3Jinan University, China 4Chinese PLA General Hospital, China |
| Pseudocode | Yes | The detailed algorithm of Pre GC is shown in Algorithm 1. |
| Open Source Code | Yes | Our code and condensed graph data regarding the testing phase are included in the supplementary materials. We will release our complete code as soon. |
| Open Datasets | Yes | To comprehensively evaluate the condensation performance of Pre GC, five graph datasets (Cora, Citeseer, Pubmed [26], OGB-Arxiv [60], and H&M [2]) are utilized in experiments. |
| Dataset Splits | Yes | Unless otherwise noted, we use the default training/validation/testing ratios to evaluate the performance of the condensed graphs for supervised tasks. ... For link prediction task, we follow [42] and randomly select 10% ratio of edges as validation and test sets, respectively. |
| Hardware Specification | Yes | The experiments are conducted on the machine with Intel(R) Xeon(R) Silver 4314 CPU @ 2.40GHz, and NVIDIA Ge Force RTX 4090 with 24GB memory and CUDA 12.0. The operating system is Ubuntu 20.04.6 with 384GB memory. |
| Software Dependencies | Yes | The experiments are conducted on the machine with Intel(R) Xeon(R) Silver 4314 CPU @ 2.40GHz, and NVIDIA Ge Force RTX 4090 with 24GB memory and CUDA 12.0. The operating system is Ubuntu 20.04.6 with 384GB memory. |
| Experiment Setup | Yes | Concretely, the smoothing factor in APPNP, SSG, and GPR. is uniformly set to 0.1, the numer of attention heads in GAT is set to 4. For GNNs [27, 65, 23, 8] that can capture long-range dependencies, we set the number of propagation layers K = 5 by default and K = 2 otherwise. For other hyperparameters, we use the combination of parameters provided by Py G by default. ... For all GNNs, the hidden unit is set to 64, with a learning rate of 0.01, weight decay of 0.0005 and the Adam optimizer employed. ... For our proposed Pre GC, we set the importance coefficient ฮณ = 0.5 in Eq. (8) by default, ฮดtmin = 0.1 for all datasets. |