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..

Robust Graph Condensation via Classification Complexity Mitigation

Authors: Jiayi Luo, Qingyun Sun, Beining Yang, Haonan Yuan, Xingcheng Fu, Yanbiao Ma, Jianxin Li, Philip S Yu

NeurIPS 2025 | Venue PDF | LLM Run Details

Reproducibility Variable Result LLM Response
Research Type Experimental Extensive experiments demonstrate the robustness of MRGC across diverse attack scenarios.
Researcher Affiliation Academia 1SKLCCSE, School of Computer Science and Engineering, Beihang University 2Laboratory for Foundations of Computer Science, University of Edinburgh 3Key Lab of Education Blockchain and Intelligent Technology, Guangxi Normal University 4Gaoling School of Artificial Intelligence, Renmin University of China 5Department of Computer Science, University of Illinois, Chicago
Pseudocode Yes Algorithm 1 The overall training pipeline of MRGC.
Open Source Code Yes 2Our code is available at https://github.com/RingBDStack/MRGC
Open Datasets Yes Datasets. We evaluate MRGC 2and the baselines on five real-world node classification datasets in a transductive setting: Cora [66], Cite Seer [66], Pub Med [66], DBLP [6], and Ogbn-arxiv [27].
Dataset Splits Yes The data split configuration follows that of [20] for the Cora, Cite Seer, Pub Med, and Ogbn-arxiv datasets. For the DBLP dataset, we use the settings from [28], performing random splits with 20 labeled nodes per class for training, 30 per class for validation, and the remaining nodes for testing.
Hardware Specification Yes All the experiments are conducted in a single NVIDIA Ge Force RTX 3090 24GB GPU.
Software Dependencies No The paper mentions GCond as a backbone, but does not specify versions for other key software components like Python, PyTorch, or CUDA.
Experiment Setup Yes The hyperparameters α, β, and γ are determined through a grid search from 1e-3 to 1e2 with logarithmic steps of 5. Details can be found in Appendix C and our code. Table 4: Hyperparameters Setting. Dataset Ratio α β γ k lr(feat) lr(adj) Cora 1.30% 1e2 1e-1 1e0 3 0.01 0.0001 2.60% 5e1 1e-1 1e0 3 0.01 0.0001 5.20% 5e1 1e-2 1e-1 3 0.01 0.0001