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..

Sparse Optimistic Information Directed Sampling

Authors: Ludovic Schwartz, Hamish Flynn, Gergely Neu

NeurIPS 2025 | Venue PDF | LLM Run Details

Reproducibility Variable Result LLM Response
Research Type Experimental We empirically demonstrate the good performance of SOIDS. 6 Experiments
Researcher Affiliation Academia Ludovic Schwartz Hamish Flynn Gergely Neu Universitat Pompeu Fabra, Barcelona, Spain EMAIL
Pseudocode No The paper describes the SOIDS algorithm and its implementation details in Section 3 and Appendix J, but no explicit pseudocode or algorithm block is provided.
Open Source Code Yes Question: Does the paper provide open access to the data and code, with sufficient instructions to faithfully reproduce the main experimental results, as described in supplemental material? Answer: [Yes] Justification: The code is available online.
Open Datasets No The action set consists of 200 random draws from the uniform distribution on [-1, 1]d. This describes how the data was generated, not providing access to an existing open dataset.
Dataset Splits No The paper describes the generation of the problem instance and action set, but as an online decision-making problem (bandits), it does not involve explicit training/testing/validation dataset splits.
Hardware Specification Yes Question: For each experiment, does the paper provide sufficient information on the computer resources (type of compute workers, memory, time of execution) needed to reproduce the experiments? Answer: [Yes] Justification: All of the experiments can be run on the CPU of a laptop.
Software Dependencies No The paper describes implementation details and specific distributions used (Laplace, Gaussian) but does not provide specific software dependencies with version numbers (e.g., Python, PyTorch, scikit-learn versions).
Experiment Setup Yes In our experiments, we always use ρ1 = 10, ρ0 = 0.1 and β = 0.1. Also, we set the learning rates to η = 1/2 and λt = min(1/2, max(sqrt(3Ct+1 / (128d(t+1))), 1/(Ct+1 * Cmin * (t+1) * (s^(2/3))))) with Ct = 5 + 2s log(edt/s)... For the ESTC baseline, we set the exploration length T1 to 50 when d = 20, 100 when d = 40 and d = 100... Also for ESTC, we set the LASSO regularisation parameter to λ = 4 * sqrt(log(d)/T1).