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..
Fast Non-Log-Concave Sampling under Nonconvex Equality and Inequality Constraints with Landing
Authors: Kijung Jeon, Michael Muehlebach, Molei Tao
NeurIPS 2025 | Venue PDF | LLM Run Details
| Reproducibility Variable | Result | LLM Response |
|---|---|---|
| Research Type | Experimental | 5 Experiments To demonstrate the sampling accuracy and efficiency, we compare OLLA and its Hutchinsonaccelerated variant (OLLA-H) against three standard constrained samplers: CLangevin [20], CHMC [20], and CGHMC [21]. |
| Researcher Affiliation | Academia | Kijung Jeon Georgia Institute of Technology EMAIL Michael Muehlebach MPI-IS EMAIL Molei Tao Georgia Institute of Technology EMAIL |
| Pseudocode | Yes | Algorithm 1 Euler Maruyama discretization of OLLA & OLLA-H |
| Open Source Code | Yes | All code is provided in the following repository: https://github.com/KraitGit/OLLA |
| Open Datasets | Yes | We evaluate the samplers on a high-dimensional Bayesian logistic regression task using a two-layer neural network trained on the German Credit dataset [41]. [41] Hans Hofmann. Statlog (German Credit Data). UCI Machine Learning Repository, 1994. DOI: https://doi.org/10.24432/C5NC77. 10, 61 |
| Dataset Splits | No | The potential function is the negative log-posterior f(v) = (log P(D|θ) + log P(θ)), where D is the training data, P(D|θ) is the log-likelihood using the sigmoid of the logits, and log P(θ) is the log-prior based on an isotropic Gaussian distribution with precision 10 3. The paper mentions "training data" but does not explicitly provide information on how the dataset was split into training, validation, or test sets with percentages or sample counts. |
| Hardware Specification | Yes | The first two experiments were executed on a desktop with an AMD Ryzen 9 7900X CPU (12 cores) with 32 GB RAM. [...] These two experiments were executed on a Linux machine equipped with Intel Xeon Gold 6226 CPU (24 cores) and 192 GB RAM. |
| Software Dependencies | No | Runs were implemented in WSL2 (Ubuntu) environment (CPU-only), using the Python and the Py Torch [54] framework. The paper mentions Python and PyTorch as software frameworks but does not provide specific version numbers for either. |
| Experiment Setup | Yes | Table 6: Hyperparameter settings for 2D synthetic examples ( t = 5 10 4) |