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 [1].
Fixed Budget Best Arm Identification in Unimodal Bandits
Authors: Debamita Ghosh, Manjesh Kumar Hanawal, Nikola Zlatanov
TMLR 2024 | Venue PDF | LLM Run Details
| Reproducibility Variable | Result | LLM Response |
|---|---|---|
| Research Type | Experimental | 7 Simulation Results In this section, we corroborate the theoretical guarantee of FB-BAUB by applying it to problem instances of varying difficulty. ... Our simulations are averaged over 1000 runs and are shown with confidence intervals. |
| Researcher Affiliation | Academia | Debamita Ghosh EMAIL IITB-Monash Research Academy Indian Institute of Technology Bombay, India. Manjesh K. Hanawal EMAIL Department of Industrial Engineering & Operations Research Indian Institute of Technology Bombay, India. Nikola Zlatanov EMAIL Faculty of Computer and Engineering Sciences Innopolis University, Russia |
| Pseudocode | Yes | ALGO 1: Fixed Budget Best Arm in Unimodal Bandits (FB-BAUB) |
| Open Source Code | Yes | The source code is available at https://github.com/ debamita-ghosh/FBBAUB. |
| Open Datasets | No | The paper describes the generation of synthetic data for simulation experiments based on specific mathematical models (Gaussian arms with defined mean structures), but it does not use or provide access to any pre-existing publicly available dataset or release the generated data for external access. The context is simulation with defined parameters, not external dataset access. |
| Dataset Splits | No | The paper uses synthetic data generated based on specific models (Gaussian arms with defined mean structures) for simulations. The concept of traditional training/test/validation splits, as used in supervised learning, does not directly apply to the multi-armed bandit problem described, where arms are sampled sequentially. |
| Hardware Specification | No | No specific hardware details (like GPU/CPU models or cloud resources) were mentioned in the paper for running the simulations. |
| Software Dependencies | No | The paper mentions simulations and provides source code but does not specify any software dependencies with version numbers (e.g., Python, PyTorch, specific libraries). |
| Experiment Setup | Yes | We consider K Gaussian arms with known variances ฯ2 i = ฯ2 = 5, assuming that the mean of the best arm is ยตk = 252. Our simulations are averaged over 1000 runs and are shown with confidence intervals. Experiment 1: ยต1 = 312, ยตk = 252, ยตK = 311, ยต2:k 1 = ยต1 + 2(k 1)(ยตk ยต1) k 1 , and ยตk +1:K = ยตk 2(k k )(ยตk ยตK). Experiment 2: ยต1 = 312, ยตk = 252, ยตK = 311, ยต2:k 1 = ยต1 + (k 1)(ยตk ยต1) (k 1) , and ยตk +1:K = ยตk (k k )(ยตk ยตK). Experiment 3: ยต1 = 0.7 and ยต2:K = ยต1 0.6(i 1). Experiment 4: ยต1 = 0.7 and ยต2:K = ยต1 0.01 1 + 4. |