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..

Fractional Langevin Dynamics for Combinatorial Optimization via Polynomial-Time Escape

Authors: Shiyue Wang, Ziao Guo, Changhong Lu, Junchi Yan

NeurIPS 2025 | Venue PDF | LLM Run Details

Reproducibility Variable Result LLM Response
Research Type Experimental Experimental results on the Maximum Independent Set, Maximum Clique, and Maximum Cut problems demonstrate that incorporating FLD advances both sampling-based and data-driven approaches, achieving state-of-the-art (SOTA) performance in most of the experiments.
Researcher Affiliation Academia 1School of Mathematical Sciences, Key Laboratory of MEA, and Shanghai Key Laboratory of PMMP, East China Normal University 2School of Computer Science & School of Artificial Intelligence, Shanghai Jiao Tong University
Pseudocode Yes Algorithm 1 FLD-EG Algorithm 2 FLD-IG (Training)
Open Source Code Yes The codes are publicly available at https://github.com/Thinklab-SJTU/FLD4CO.
Open Datasets Yes Datasets. (1) MIS datasets: Following the benchmarks in [17], we evaluate our algorithms on two graph classes: Revised Model B (RB) instances [66] and Erd os Rényi (ER) random graphs [16] with node weight set to 1; (2) Maximum Clique dataset: we use the single RB graph which is introduced in MIS datasets for the evaluation; (3) Max Cut dataset: we compare our algorithms with the baselines on the Barabási-Albert (BA) graphs [3].
Dataset Splits Yes The size of the training set and the validation set is separately 1000 and 500 graphs for all datasets except for ER-[9000, 11000] (that is, the ER graphs contain 9000 to 11000 nodes), and the test size is 500 for RB and BA graphs; 128 for ER-[700-800] and 16 for ER-[9000, 11000].
Hardware Specification Yes Experiments are conducted on a Linux workstation using an H100 GPU and an Intel(R) Xeon(R) Platinum 8468 CPU, with programs implemented in Py Torch.
Software Dependencies No The paper mentions "programs implemented in Py Torch." but does not provide specific version numbers for PyTorch or other libraries, which is required for a reproducible software description.
Experiment Setup Yes We show the utilized hyperparameter values of FLD-EG and FLD-IG in Table 3 and Table 4, respectively.