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].
Bandit Convex Optimization in Non-stationary Environments
Authors: Peng Zhao, Guanghui Wang, Lijun Zhang, Zhi-Hua Zhou
JMLR 2021 | Venue PDF | LLM Run Details
| Reproducibility Variable | Result | LLM Response |
|---|---|---|
| Research Type | Experimental | Despite that the focus of this paper is mainly theoretical, we conduct empirical evaluations on several datasets to validate the effectiveness of our proposed approach. Finally, we present empirical studies to validate the effectiveness of the proposed approach. |
| Researcher Affiliation | Academia | Peng Zhao EMAIL Guanghui Wang EMAIL Lijun Zhang EMAIL Zhi-Hua Zhou EMAIL National Key Laboratory for Novel Software Technology Nanjing University, Nanjing 210023, China |
| Pseudocode | Yes | Algorithm 1 Bandit Gradient Descent (BGD) ... Algorithm 2 PBGD: Meta-algorithm ... Algorithm 3 MABCO: Meta-algorithm ... Algorithm 4 MABCO: Expert-algorithm |
| Open Source Code | No | The paper does not explicitly state that source code for the described methodology is publicly available, nor does it provide a link to a code repository. |
| Open Datasets | Yes | The dataset can be downloaded from https://www.win.tue.nl/~mpechen/data/Drift Sets/. The datasets can be downloaded from https://home.isr.uc.pt/~fasouza/datasets.html. |
| Dataset Splits | No | The paper describes generating synthetic datasets and using real-world datasets in an online learning scenario, where the player makes predictions sequentially and the performance is measured by cumulative loss. It does not specify explicit training/test/validation splits in the traditional sense. |
| Hardware Specification | No | The paper does not provide specific hardware details such as GPU models, CPU types, or memory specifications used for running the experiments. |
| Software Dependencies | No | The paper does not provide specific software dependency details, such as library names with version numbers, needed to replicate the experiments. |
| Experiment Setup | Yes | The player will adopt a simple linear model, denoted by byt = g(wt, xt) = w T t xt. Besides, the loss function is chosen as the squared loss â„“(y, by) = (y by)2. Therefore, the online function at each iteration ft : W 7 R is essentially a couple of loss function â„“and the instance (xt, yt), namely, ft(w) = â„“(w Txt, yt)2. Moreover, the features of the datasets are normalized to the range of [0, 1], which implies the diameter of the feasible set R = 1. the step size of BGD algorithm is set to R2/(d T 3/4) and R/(d L T) for one-point and two-point BCO respectively, as suggested by their theory. |