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..
Non-Stationary Lipschitz Bandits
Authors: Nicolas Nguyen, Solenne Gaucher, Claire Vernade
NeurIPS 2025 | Venue PDF | LLM Run Details
| Reproducibility Variable | Result | LLM Response |
|---|---|---|
| Research Type | Experimental | While our work is mainly theoretical, we also analyze the computational worst-case complexity of MDBE in Appendix H and show its empirical performance on a synthetic example in Appendix I. In this section, we illustrate some numerical experiments to show the empirical performances of MDBE on a synthetic dataset. |
| Researcher Affiliation | Academia | Nicolas Nguyen University of Tübingen EMAIL Solenne Gaucher École Polytechnique, CMAP EMAIL Claire Vernade University of Tübingen EMAIL |
| Pseudocode | Yes | For completeness, the full pseudocode of MDBE is provided in Appendix B, and all detailed regret proofs can be found in Appendix D. Algorithm: MDBE: Multi-Depth Bin Elimination |
| Open Source Code | Yes | The code for these implementations is available at https://github.com/ nguyenicolas/NS_Lipschitz_Bandits. |
| Open Datasets | No | Environment. We simulate a 1-Lipschitz, piecewise-linear reward function defined over the action space [0, 1], with a single peak shifting smoothly from x = 0.3 to x = 0.7 every 105 rounds. This setup induces LT = 10 significant shifts over a time horizon T = 106. (Explanation: The paper uses a synthetic dataset generated for the simulation, and does not provide access information for a publicly available or open dataset.) |
| Dataset Splits | No | Environment. We simulate a 1-Lipschitz, piecewise-linear reward function defined over the action space [0, 1], with a single peak shifting smoothly from x = 0.3 to x = 0.7 every 105 rounds. This setup induces LT = 10 significant shifts over a time horizon T = 106. (Explanation: The paper describes a simulation environment for a bandit problem, not a dataset with train/test/validation splits for supervised learning.) |
| Hardware Specification | No | No particular computer resource is needed to run the experiments. (Explanation: The paper explicitly states that no particular computer resource is needed, implying no specific hardware details are provided.) |
| Software Dependencies | No | The code for these implementations is available at https://github.com/ nguyenicolas/NS_Lipschitz_Bandits. (Explanation: While the paper provides a link to the code, it does not explicitly list specific software dependencies with version numbers within the text.) |
| Experiment Setup | Yes | Environment. We simulate a 1-Lipschitz, piecewise-linear reward function defined over the action space [0, 1], with a single peak shifting smoothly from x = 0.3 to x = 0.7 every 105 rounds. This setup induces LT = 10 significant shifts over a time horizon T = 106. Thus, the mean reward changes every round, but only ten of these changes are significant under our framework. ... Results are averaged over 100 independent runs, with 95% confidence intervals of the mean dynamic regret shown. |