Factored-Reward Bandits with Intermediate Observations

Authors: Marco Mussi, Simone Drago, Marcello Restelli, Alberto Maria Metelli

ICML 2024 | Conference PDF | Archive PDF | Plain Text | LLM Run Details

Reproducibility Variable Result LLM Response
Research Type Experimental Numerical simulations are provided in Appendix E.
Researcher Affiliation Academia 1Politecnico di Milano, Milan, Italy.
Pseudocode Yes Algorithm 1: F-UCB.
Open Source Code Yes The code of the experiments can be found at https://github.com/marcomussi/FRB.
Open Datasets No The paper generates synthetic data for its experiments, describing the process: 'We draw the expected values µi,j for i P Jd K and j P Jk K from a uniform distribution in the range r0.7, 1s.' It does not use a pre-existing public dataset or provide specific access information for the generated data.
Dataset Splits No The paper describes experimental settings but does not specify explicit training, validation, or test dataset splits (e.g., percentages or sample counts). It refers to the 'learning horizon T' but not data partitioning.
Hardware Specification No The paper does not provide any specific hardware details such as GPU/CPU models, memory, or cloud computing resources used for running the experiments.
Software Dependencies No The paper describes algorithms and numerical simulations but does not list any specific software dependencies with version numbers (e.g., Python 3.x, PyTorch 1.x).
Experiment Setup Yes Setting For the sake of simplicity in the presentation of the results, we consider the scenario in which all the problem dimensions present the same number of actions (i.e., k1 kd : k). Moreover, we consider the setting in which the intermediate observations are drawn from Gaussian distributions with mean µi,aiptq for every action component aiptq in position i of the action vector a, formally xiptq Npµi,aiptq, σ2q, @i P Jd K. We consider values of k P J3, 5K, and values of d P J4K. We draw the expected values µi,j for i P Jd K and j P Jk K from a uniform distribution in the range r0.7, 1s. We fix a value of σ 0.1. ... We evaluate the performances in terms of cumulative regret with T 104, averaged over 50 trials.