No-Regret Exploration in Goal-Oriented Reinforcement Learning

Authors: Jean Tarbouriech, Evrard Garcelon, Michal Valko, Matteo Pirotta, Alessandro Lazaric

ICML 2020 | Conference PDF | Archive PDF | Plain Text | LLM Run Details

Reproducibility Variable Result LLM Response
Research Type Experimental We introduce UC-SSP, the first no-regret algorithm in this setting, and prove a regret bound scaling as e O(DS ADK) after K episodes for any unknown SSP with S states, A actions, positive costs and SSP-diameter D... Finally, we support our theoretical findings with experiments in App. J.
Researcher Affiliation Collaboration 1Facebook AI Research, Paris, France 2Seque L team, Inria Lille Nord Europe, France.
Pseudocode Yes Algorithm 1 UC-SSP algorithm and Algorithm 2 EVISSP
Open Source Code No The paper does not contain an explicit statement about releasing source code for the described methodology, nor does it provide a link to a code repository.
Open Datasets No The paper describes using custom 'gridworld environments' for experiments, but it does not provide concrete access information (link, DOI, formal citation) for a publicly available or open dataset.
Dataset Splits No The paper does not provide specific dataset split information (exact percentages, sample counts, or detailed splitting methodology) for training, validation, and test sets.
Hardware Specification No The paper does not provide specific hardware details (exact GPU/CPU models, processor types, or 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 experiment.
Experiment Setup Yes We report our experimental results in App. J. for two different environments described in Fig. 2 (a) and Fig. 2 (b). For the parameters of UC-SSP, we set the confidence δ = 0.05 and use cmin = 1 and cmax = 10 for the general SSP case, as well as cmin = cmax = 1 for the uniform-cost SSP case. We average the regret over 50 independent runs and plot the average regret with 95% confidence intervals. For the discount factor γ of UCRL2 and UCRL2B, we choose γ = 0.95. For UCBVI, we use H = 100 as the fixed horizon.