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..

Memorization in Graph Neural Networks

Authors: Adarsh Jamadandi, Jing Xu, Adam Dziedzic, Franziska Boenisch

NeurIPS 2025 | Venue PDF | LLM Run Details

Reproducibility Variable Result LLM Response
Research Type Experimental Overall, we perform extensive experiments on various real-world datasets and semi-synthetic datasets [68] with various GNN backbones such as SGC [65], GCN [37], Graph SAGE [28], GATv2 [13] and Graph Transformer (GT) to confirm our findings.
Researcher Affiliation Academia CNRS IRISA1, CISPA Helmholtz Center for Information Security2
Pseudocode No The paper does not contain a clearly labeled pseudocode or algorithm block.
Open Source Code Yes Answer: [Yes] Justification: We provide all the code necessary to reproduce our results. The datasets are publicly available.
Open Datasets Yes Overall, we perform extensive experiments on various real-world datasets and semi-synthetic datasets [68] with various GNN backbones such as SGC [65], GCN [37], Graph SAGE [28], GATv2 [13] and Graph Transformer (GT) to confirm our findings. Answer: [Yes] Justification: We provide all the code necessary to reproduce our results. The datasets are publicly available.
Dataset Splits Yes We divide our datasets into 60%/20%/20% as training, testing, and validation sets, respectively. Out of the 60% training set S, we further randomly divide the datasets into three disjoint subsets with ratios SS = 50%, SC = 25%, SI = 25%.
Hardware Specification Yes All of our experiments were done on a T4 GPU highlighting the computationally efficiency of our proposed framework.
Software Dependencies No The paper mentions "Py Torch Geometric [23] for all of our experiments" and "Utilizing optimized libraries like scikit-learn", but does not provide specific version numbers for these software components.
Experiment Setup Yes We document all the necessary hyperparameters for reproducing the results in the Appendix H. Table 18: Hyperparameters and dataset statistics. Dataset Hidden Dimension LR #Layers |V| |E| # Classes Homophily Level Node Label Informativeness Cora 32 0.01 3 2708 10138 7 0.7637 0.5763