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..

BrainFlow: A Holistic Pathway of Dynamic Neural System on Manifold

Authors: Zhixuan Zhou, Tingting Dan, Guorong Wu

NeurIPS 2025 | Venue PDF | LLM Run Details

Reproducibility Variable Result LLM Response
Research Type Experimental In this section, we conduct extensive experiments on synthetic datasets and real brain structural-functional network datasets for validating the effectiveness of Brain Flow.
Researcher Affiliation Academia Departments of Computer Science and Psychiatry University of North Carolina at Chapel Hill Chapel Hill, NC 27599 EMAIL;EMAIL
Pseudocode Yes Algorithm 1 Training of Brain Flow
Open Source Code No The paper states in its NeurIPS checklist that it provides open access to data and code ('Answer: [Yes] Justification: See Section 5.'). However, Section 5 and Appendix D.1, which discusses implementation, only state 'We base our implementation on [30]' without providing a direct link to the code developed specifically for this paper or an unambiguous statement of its release.
Open Datasets Yes We compare our methods to baselines and large-scale neuroimaging datasets from UK Biobank and Human Connectome Project. In addition to competitive or state-of-the-art performance on different metrics, our method shows faster training speed and more stable inference. The detailed implementation and experiment settings are shown in Appendix D and the data description of Human Connectome Project-Aging (HCP-A) [26], HCP-Young Adult (HCP-YA) [27], UK Biobank [28] are listed in Appendix D.4
Dataset Splits Yes The datasets are split into train, validation, test set with ratio 7:1:2.
Hardware Specification Yes We run all the experiments using a single NVIDIA A6000 GPU.
Software Dependencies No The flow-based approach uses an 8-layer Transformer [31]... For inference, we employ an Euler solver... All the models are trained with 1000 epoch using Adam W [32] with Cosine Annealing learning rate schedule [33]. No specific version numbers are provided for these software components.
Experiment Setup Yes All the models are trained with 1000 epoch using Adam W [32] with Cosine Annealing learning rate schedule [33]. We do grid search on the following hyperparameters: (1) learning rate: {0.001, 0.0005, 0.0001} (2) Layer of Model: {4, 6, 8, 12} and (3) Batch size: {256, 512, 1024}.