Lipschitzness is all you need to tame off-policy generative adversarial imitation learning.

Deep learning Generative adversarial networks Imitation learning Lipschitz-continuity Reinforcement learning

Journal

Machine learning
ISSN: 0885-6125
Titre abrégé: Mach Learn
Pays: United States
ID NLM: 9881780

Informations de publication

Date de publication:
2022
Historique:
received: 01 08 2020
revised: 18 01 2022
accepted: 27 01 2022
entrez: 23 5 2022
pubmed: 24 5 2022
medline: 24 5 2022
Statut: ppublish

Résumé

Despite the recent success of reinforcement learning in various domains, these approaches remain, for the most part, deterringly sensitive to hyper-parameters and are often riddled with essential engineering feats allowing their success. We consider the case of off-policy generative adversarial imitation learning, and perform an in-depth review, qualitative and quantitative, of the method. We show that forcing the learned reward function to be local Lipschitz-continuous is a

Identifiants

pubmed: 35602587
doi: 10.1007/s10994-022-06144-5
pii: 6144
pmc: PMC9114147
doi:

Types de publication

Journal Article

Langues

eng

Pagination

1431-1521

Informations de copyright

© The Author(s) 2022.

Déclaration de conflit d'intérêts

Conflict of interestThe authors declare that they have no competing interests.

Références

Nature. 2016 Jan 28;529(7587):484-9
pubmed: 26819042
Ann Oper Res. 2013 Sep 1;208(1):383-416
pubmed: 24049244
Nature. 2015 Feb 26;518(7540):529-33
pubmed: 25719670
IEEE Trans Neural Netw. 1994;5(3):363-71
pubmed: 18267804
Neural Comput. 1997 Jan 1;9(1):1-42
pubmed: 9117894
Nature. 2019 Nov;575(7782):350-354
pubmed: 31666705

Auteurs

Lionel Blondé (L)

University of Geneva, Geneva School of Business Administration - HES-SO, Geneva, Switzerland.

Pablo Strasser (P)

University of Geneva, Geneva School of Business Administration - HES-SO, Geneva, Switzerland.

Alexandros Kalousis (A)

University of Geneva, Geneva School of Business Administration - HES-SO, Geneva, Switzerland.

Classifications MeSH