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..
Balancing Risk and Reward: A Batched-Bandit Strategy for Automated Phased Release
Authors: Yufan Li, Jialiang Mao, Iavor Bojinov
NeurIPS 2023 | Venue PDF | LLM Run Details
| Reproducibility Variable | Result | LLM Response |
|---|---|---|
| Research Type | Experimental | 4 Numerical and empirical experiments |
| Researcher Affiliation | Collaboration | Yufan Li1, Jialiang Mao2, Iavor Bojinov3 1Harvard University 2Linked In Corporation 3Harvard Business School |
| Pseudocode | Yes | Algorithm 1 Output ramp size adaptively |
| Open Source Code | No | The paper does not provide a direct link to source code or an explicit statement about its release. |
| Open Datasets | No | The paper uses 'semi-real Linked In ramp schedule comparison' where data 'is simulated from (4) using stage-wise µtrue (w), σ(w)2, w = 0, 1 (both unobserved)' and states 'Due to privacy constraints, the individual-level data is not available'. No concrete access information for a publicly available dataset is provided. |
| Dataset Splits | No | The paper describes a sequential A/B testing approach and simulations, but does not provide details on specific training/validation/test dataset splits like percentages or sample counts. |
| Hardware Specification | No | The paper mentions that 'Nt are incoming population size reduced by 104 factor for tractability on a personal computer', but does not provide specific hardware details like CPU/GPU models or memory amounts used for running experiments. |
| Software Dependencies | No | The paper does not specify any software dependencies with version numbers (e.g., programming languages, libraries, or frameworks). |
| Experiment Setup | Yes | For each scenario, we set T = 10 with Nt = 500, t and we choose non-informative prior µ0(w) = 0, σ0(w)2 = 100, w = 0, 1. We assume model variance is known; however, using (6) to estimate the variance gives similar results. and We set B = 500 to produce (h), although the model is not budget-aware. and Under the legends (B, δ) , we set bt = B, t = 1 (1 δ)1/T , t. We also use (i) ration budget to denote (B, δ) = ( 500, 0.01), bt = 400, t 5, bt = 500, t > 5 and t = 1 (1 δ)1/T , t; (ii) ration tolerance to denote (B, δ) = ( 500, 0.01), bt = 500, t and t = 0.0001, t 5, t = 0.0019, t > 5 |