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..
Adjusted Count Quantification Learning on Graphs
Authors: Clemens Damke, Eyke Hüllermeier
NeurIPS 2025 | Venue PDF | LLM Run Details
| Reproducibility Variable | Result | LLM Response |
|---|---|---|
| Research Type | Experimental | We assess the performance of SIS and NACC on a series of graph quantification tasks using, both, PPS and covariate shift. The quantifiers are applied to the predictions of a set of node classifiers. As a baseline we compare our proposed GQL methods with MLPE, (P)CC and (P)ACC. We build upon the Qua Py Python library (BSD 3-Clause). Further details can be found in Appendix A. |
| Researcher Affiliation | Academia | Clemens Damke LMU Munich, MCML EMAIL Eyke H ullermeier LMU Munich, MCML, DFKI EMAIL |
| Pseudocode | No | The paper does not contain any sections explicitly labeled 'Pseudocode' or 'Algorithm'. |
| Open Source Code | Yes | Our code is available at https://github.com/Cortys/ graph-quantification; it includes a versioned list of all dependencies that were used. |
| Open Datasets | Yes | For this reason, we synthetically generate quantification tasks from the following five node classification datasets: 1. Cora ML, 2. Cite Seer, 3. Pub Med, 4. Amazon Photos and 5. Amazon Computers [21, 14, 13, 31, 25, 20, 32]. Additionally, we conduct a study on the Twitch Gamers dataset [28]; |
| Dataset Splits | Yes | All reported results were obtained by averaging over 10 random splits of the node set into classifier-train/quantifier-train/test, with sizes of 5%/15%/80% respectively. |
| Hardware Specification | Yes | All experiments were conducted on a single machine with an AMD Ryzen 9 5950X CPU, 64GB RAM and an Nvidia RTX 4090 GPU with 24GB VRAM. |
| Software Dependencies | No | In our implementation, we use torch-geometric [7] (MIT license) for the graph neural network (GNN) models, while the Qua Py Python library (BSD 3-Clause) was used for the quantification methods. Additionally, we used Nvidia s cu Graph library (Apache 2.0 license) for GPU-based graph traversal and distance computation, e.g., to create BFS-based covariate shift. Our code is available at https://github.com/Cortys/ graph-quantification; it includes a versioned list of all dependencies that were used. |
| Experiment Setup | Yes | All models are trained using the same training splits and hyperparameters, and two hidden layers/convolutions with widths of 64 and ReLU activations. Each model is trained ten times on each of the ten splits per dataset, totalling 100 models per dataset, with which each quantifier is evaluated. |