000 03886naa a22004937i 4500
001 UMUN-OA-20261005-AI-046
003 local
005 20261005174826.0
006 m o d
007 cr||||||||||||
008 261005s2023 xx o u000 0 eng d
024 7 _a10.1109/tnnls.2023.3250269
_2doi
035 _a(local)OpenAlex:W4360584316
040 _alocal
_beng
_clocal
041 0 _aeng
100 0 _aRafael Figueiredo Prudencio
_eauthor.
245 1 2 _aA Survey on Offline Reinforcement Learning: Taxonomy, Review, and Open Problems /
_cRafael Figueiredo Prudencio; Marcos R. O. A. Maximo; Esther Luna Colombini.
264 1 _a[Place of publication not identified] :
_bInstitute of Electrical and Electronics Engineers,
_c2023.
300 _a1 online resource (1 article).
336 _atext
_btxt
_2rdacontent
337 _acomputer
_bc
_2rdamedia
338 _aonline resource
_bcr
_2rdacarrier
500 _aPublication type: Journal article.
500 _aLinked open-access copy: published version.
506 0 _aOpen access; no subscription required for the linked open-access version.
_fUnrestricted online access
_2star
520 3 _aWith the widespread adoption of deep learning, reinforcement learning (RL) has experienced a dramatic increase in popularity, scaling to previously intractable problems, such as playing complex games from pixel observations, sustaining conversations with humans, and controlling robotic agents. However, there is still a wide range of domains inaccessible to RL due to the high cost and danger of interacting with the environment. Offline RL is a paradigm that learns exclusively from static datasets of previously collected interactions, making it feasible to extract policies from large and diverse training datasets. Effective offline RL algorithms have a much wider range of applications than online RL, being particularly appealing for real-world applications, such as education, healthcare, and robotics. In this work, we contribute with a unifying taxonomy to classify offline RL methods. Furthermore, we provide a comprehensive review of the latest algorithmic breakthroughs in the field using a unified notation as well as a review of existing benchmarks' properties and shortcomings. Additionally, we provide a figure that summarizes the performance of each method and class of methods on different dataset properties, equipping researchers with the tools to decide which type of algorithm is best suited for the problem at hand and identify which classes of algorithms look the most promising. Finally, we provide our perspective on open problems and propose future research directions for this rapidly growing field.
540 _aCreative Commons Attribution (licence family recorded by the source; version not supplied in metadata).
_uhttps://openalex.org/W4360584316
546 _aIn English.
588 0 _aDescription generated from OpenAlex bibliographic metadata retrieved on 5 October 2026; full text linked from the source.
590 _aCourse allocation: B. Sc. Artificial Intelligence.
650 4 _aoffline reinforcement learning
650 4 _abenchmark dataset
650 4 _aalgorithm comparison
653 _aB. Sc. Artificial Intelligence
653 _aUMUN open access collection
653 _areinforcement learning
655 4 _aJournal articles.
700 0 _aMarcos R. O. A. Maximo
_eauthor.
700 0 _aEsther Luna Colombini
_eauthor.
773 0 _tIEEE Transactions on Neural Networks and Learning Systems
_gVol. 35, no. 8, (2023), pages/article 10237-10257
_x2162-237X
856 4 0 _uhttps://ieeexplore.ieee.org/ielx7/5962385/6104215/10078377.pdf
_yRead open-access publication
_zOpen access.
_70
_qapplication/pdf
884 _aCourse selection and automated MARC21 enrichment from OpenAlex metadata.
_g20261005
_khttps://openalex.org/W4360584316
_qlocal
942 _n0
999 _c603
_d603