Notice: This website is continuously updated as new conference proceedings and journal volumes are published. Consequently, the figures presented here may differ from those reported in our publications.
A brief overview of our methodology is provided on the Methods page, along with references to publications documenting our approach and links to the code and data used in each study.
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.
Exploration via Feature Perturbation in Contextual Bandits
Authors: Seouh-won Yi, Min-hwan Oh
NeurIPS 2025 | Venue PDF | LLM Run Details | Input Tokens: 36,463 Total number of tokens sent to the LLM as input for this paper's analysis. | Output Tokens: 3,893 Total number of tokens produced by the LLM (including reasoning/thinking tokens) for this paper's analysis.
| Reproducibility Variable | Result | LLM Response |
|---|---|---|
| Research Type | Experimental | We conduct experiments in two contextual bandit settings: (i) generalized linear bandits (GLBs), including linear and logistic models, and (ii) nonlinear contextual bandits based on neural networks. In each setting, we compare the proposed method with state-of-the-art baselines across varying feature dimensions and datasets. All results are averaged over 100 independent runs to ensure robustness. Detailed experimental setups are provided in Appendix H. |
| Researcher Affiliation | Academia | Seouh-won Yi Seoul National University EMAIL Min-hwan Oh Seoul National University EMAIL |
| Pseudocode | Yes | Algorithm 1 GLM-FP: Feature Perturbation in Generalized Linear Bandits |
| Open Source Code | Yes | We have included the code in the supplementary material. |
| Open Datasets | Yes | We evaluate Deep FP on three UCI benchmark datasets: shuttle (7 classes, 9 features), isolet (26 classes, 617 features), and mushroom (binary, 112 features). |
| Dataset Splits | Yes | We consider a time-varying set of K = 100 arms per round over a horizon of T = 20,000 (linear) or T = 10,000 (logistic). Context vectors and the true parameter vector ฮธ are sampled from a standard multivariate Gaussian distribution and normalized to satisfy the boundedness assumption. In the logistic case, we use the sigmoid link function ยต(x) = 1/(1 + e x) and constrain ฮธ 4 so that the logits lie in [ 4, 4]. For the linear bandit, the reward noise is sampled from N(0, 1). |
| Hardware Specification | Yes | All GLB runs were performed on a standard CPU server equipped with an Intel Xeon Silver 4210R processor (40 threads), with each run completing within a few minutes. The neural bandit experiments were conducted using a single NVIDIA RTX 3090 GPU. |
| Software Dependencies | No | The paper mentions 'Adam optimizer' but does not specify any version numbers for software libraries, frameworks, or languages required to replicate the experiment. |
| Experiment Setup | Yes | The confidence level is set as ฮด = 1/T, and the regularization parameter is fixed at ฮป = 10^-4. |