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..
Neural Fractional Attention Differential Equations
Authors: Qiyu Kang, Wenjun Cui, Xuhao Li, Yuxin Ma, Xueyang Fu, Wee Peng Tay, Yidong Li, Zheng-Jun Zha
NeurIPS 2025 | Venue PDF | LLM Run Details
| Reproducibility Variable | Result | LLM Response |
|---|---|---|
| Research Type | Experimental | We conduct extensive experiments on fluid flow, graph learning problems, spatio-temporal traffic forecasting, urban population mobility and biological neural spike trains tasks, demonstrating that FADE consistently outperforms integer-order neural ODE models and existing fractional approaches, confirming its superior capacity for modeling complex dynamics. |
| Researcher Affiliation | Academia | 1University of Science and Technology of China 2Shanxi University 3Anhui University 4Nanyang Technological University 5Beijing Jiaotong University |
| Pseudocode | No | The paper describes methods in regular paragraph text and mathematical equations, but it does not contain a clearly labeled 'Pseudocode' or 'Algorithm' block, nor does it present structured steps in a code-like format. |
| Open Source Code | Yes | The code is available at https://github.com/cuiwjTech/NeurIPS2025_FADE. |
| Open Datasets | Yes | We evaluated the proposed FADE on turbulent boundary-layer flow [53]... For the Airport dataset, we adopt preprocessing protocols from [55]. For the others, we follow the protocols from GRAND [56]... The authors revealed critical limitations in the commonly used benchmark datasets for evaluating models on heterophilic graphs [86]. To address this, they introduced several new datasets, such as Roman-empire, Wiki-cooc, Questions, Workers and Amazon-ratings... We evaluate the effectiveness of STDDE using five real-world traffic datasets: Pe MSD7(M), Pe MSD7(L), Pe MS04, Pe MS07, and Pe MS08. These datasets are sourced from the Caltrans Performance Measurement System [59]... We tested the experimental results on spiking datasets Allen and Retina [95]. |
| Dataset Splits | Yes | For the Airport dataset, we adopt preprocessing protocols from [55]. For the others, we follow the protocols from GRAND [56], specifically using random splits applied to the largest connected component... These datasets are pre-divided into training, validation, and testing sets using a 6:2:2 ratio. |
| Hardware Specification | Yes | All implementations are developed using the Py Torch framework[52] on a single NVIDIA RTX4090 24GB GPU. |
| Software Dependencies | No | All implementations are developed using the Py Torch framework[52]... No specific version numbers for PyTorch or any other libraries are provided in the main text or supplemental material. |
| Experiment Setup | Yes | Taking the Cora dataset as an example to illustrate the experimental parameter settings, we set time = 25, step size = 1, learning rate = 0.01, weight decay = 0.05, epoch = 800 and dim = 256. For more details regarding the dataset and extended experimental results, please refer to Appendix G... Maintaining consistency with [4], we train all datasets for 200 epochs using the Adam optimizer and a batch size of 64. An early stopping strategy is applied, with a patience of 15 iterations on the validation dataset. |