Latent Bandits.

Authors: Odalric-Ambrym Maillard, Shie Mannor

ICML 2014 | Conference PDF | Archive PDF | Plain Text | LLM Run Details

Reproducibility Variable Result LLM Response
Research Type Experimental Numerical experiments show that, in the most challenging agnostic case, the proposed algorithm achieves excellent performance in several difficult scenarios.
Researcher Affiliation Academia Odalric-Ambrym Maillard ODALRIC-AMBRYM.MAILLARD@ENS-CACHAN.ORG The Technion, Faculty of Electrical Engineering 32000 Haifa, ISRAEL Shie Mannor SHIE@EE.TECHNION.AC.IL The Technion, Faculty of Electrical Engineering 32000 Haifa, ISRAEL
Pseudocode Yes Algorithm 1 The Single-K-UCB algorithm. ... Algorithm 2 The Multiple-K-UCB algorithm. ... Algorithm 3 The UCB on B algorithm ... Algorithm 4 The A-UCB algorithm
Open Source Code No The paper does not provide concrete access to source code for the methodology described. It links to an extended version of the paper for proofs.
Open Datasets No The paper describes generating data for its experiments based on Bernoulli distributions and specified parameters for |A|, |B|, |C|, and Υ(b), but it does not use a named public dataset or provide access information for a generated dataset.
Dataset Splits No The paper does not provide specific dataset split information (e.g., train/validation/test percentages or counts). It describes the parameters for the generated environments for numerical experiments.
Hardware Specification No The paper does not provide specific hardware details (e.g., exact GPU/CPU models, memory amounts) used for running its experiments.
Software Dependencies No The paper does not provide specific ancillary software details (e.g., library or solver names with version numbers) needed to replicate the experiments.
Experiment Setup Yes For each experiment, we show the number of actions |A|, of users |B|, of classes |C|, and the parameters {µa,c}a A,c C when there are not too many. We plot the regret of all algorithms on the same figure: A thick line is used for the mean regret and dashed lines for quantiles at levels 0.25, 0.5, 0.75, 0.95 and 0.99. In all experiments, the parameters {Υ(b)}b B are defined by Υ(b) = wb/ P b B wb, where the weights wb are drawn uniformly randomly in [0.1, 0.9]. Thus for each class, the distortion factor γc is less than 9, and we set the parameter γ of A-UCB to the value γ = 9. For one experiment with given fixed parameters, the algorithms are run over several trials (500) for a large time horizon N = 25000.