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..
Uncover Governing Law of Pathology Propagation Mechanism Through A Mean-Field Game
Authors: Tingting Dan, Zhihao Fan, Guorong Wu
NeurIPS 2025 | Venue PDF | LLM Run Details
| Reproducibility Variable | Result | LLM Response |
|---|---|---|
| Research Type | Experimental | Experimental results on public neuroimaging datasets demonstrate that our explainable deep model not only yields precise and reliable predictions of future tau progression for unseen new subjects but also provides a new window to uncover new understanding of pathology propagation in AD through learning-based approaches. |
| Researcher Affiliation | Academia | Tingting Dan Zhihao Fan Guorong Wu Departments of Psychiatry and Computer Science University of North Carolina at Chapel Hill Chapel Hill, NC 27599 EMAIL;EMAIL |
| Pseudocode | Yes | The pseudocode for our method is presented in Algorithm 1. The full implementation including all hyperparameter settings is available from our anonymous Git Hub repository: https://github. com/Dandy5721/MFG4AD2025. |
| Open Source Code | Yes | The full implementation including all hyperparameter settings is available from our anonymous Git Hub repository: https://github. com/Dandy5721/MFG4AD2025. |
| Open Datasets | Yes | We evaluate the performance of MFG4AD using two longitudinal tau PET datasets: the Alzheimer s Disease Neuroimaging Initiative (ADNI) [25] and the Open Access Series of Imaging Studies (OASIS) [28]. |
| Dataset Splits | Yes | For all experiments, we conduct 5-fold cross-validation. |
| Hardware Specification | Yes | All experiments were conducted on an RTX A5000 GPU. |
| Software Dependencies | No | Finally, for compatibility with graph learning frameworks such as Py Torch Geometric, we exported the graph as an edge list edge_index Z2 |E| and a corresponding edge weight vector edge_weight R|E|. All models are trained for 1,000 epochs with Adam [27] optimizer. |
| Experiment Setup | Yes | Each generator update is preceded by n C = 5 critic updates, with learning rates set to ηC = 1 10 5 for the critic and ηG = 1 10 4 for the generator. ... The generator s loss combines the adversarial term with an ℓ1 = |ˆui(t + 1) ui(t + 1)|1 reconstruction penalty weighted by λ = 10, ensuring both realistic and accurate predictions. All models are trained for 1,000 epochs with Adam [27] optimizer. |