Automated Transfer for Reinforcement Learning Tasks

Haitham Bou Ammar, S. Chen, K. Tuyls, G. Weiss

Research output: Contribution to journalArticleAcademicpeer-review

Abstract

Reinforcement learning applications are hampered by the tabula rasa approach taken by existing techniques. Transfer for reinforcement learning tackles this problem by enabling the reuse of previously learned behaviours. To be fully autonomous a transfer agent has to: (1) automatically choose a relevant source task(s) for a given target, (2) learn about the relation between the tasks, and (3) effectively and efficiently transfer between tasks. Currently, most transfer frameworks require substantial human intervention in at least one of the previous three steps. This discussion paper aims at: (1) positioning various knowledge re-use algorithms as forms of transfer, and (2) arguing the validity and possibility of autonomous transfer by detailing potential solutions to the above three steps.
Original languageEnglish
Pages (from-to)7-14
Number of pages8
JournalKünstliche Intelligenz
Volume28
Issue number1
DOIs
Publication statusPublished - 2014

Cite this

Ammar, Haitham Bou ; Chen, S. ; Tuyls, K. ; Weiss, G. / Automated Transfer for Reinforcement Learning Tasks. In: Künstliche Intelligenz. 2014 ; Vol. 28, No. 1. pp. 7-14.
@article{7679650012324ddd97d4af45181dafe3,
title = "Automated Transfer for Reinforcement Learning Tasks",
abstract = "Reinforcement learning applications are hampered by the tabula rasa approach taken by existing techniques. Transfer for reinforcement learning tackles this problem by enabling the reuse of previously learned behaviours. To be fully autonomous a transfer agent has to: (1) automatically choose a relevant source task(s) for a given target, (2) learn about the relation between the tasks, and (3) effectively and efficiently transfer between tasks. Currently, most transfer frameworks require substantial human intervention in at least one of the previous three steps. This discussion paper aims at: (1) positioning various knowledge re-use algorithms as forms of transfer, and (2) arguing the validity and possibility of autonomous transfer by detailing potential solutions to the above three steps.",
author = "Ammar, {Haitham Bou} and S. Chen and K. Tuyls and G. Weiss",
year = "2014",
doi = "10.1007/s13218-013-0286-8",
language = "English",
volume = "28",
pages = "7--14",
journal = "K{\"u}nstliche Intelligenz",
issn = "0933-1875",
publisher = "Springer Verlag",
number = "1",

}

Automated Transfer for Reinforcement Learning Tasks. / Ammar, Haitham Bou; Chen, S.; Tuyls, K.; Weiss, G.

In: Künstliche Intelligenz, Vol. 28, No. 1, 2014, p. 7-14.

Research output: Contribution to journalArticleAcademicpeer-review

TY - JOUR

T1 - Automated Transfer for Reinforcement Learning Tasks

AU - Ammar, Haitham Bou

AU - Chen, S.

AU - Tuyls, K.

AU - Weiss, G.

PY - 2014

Y1 - 2014

N2 - Reinforcement learning applications are hampered by the tabula rasa approach taken by existing techniques. Transfer for reinforcement learning tackles this problem by enabling the reuse of previously learned behaviours. To be fully autonomous a transfer agent has to: (1) automatically choose a relevant source task(s) for a given target, (2) learn about the relation between the tasks, and (3) effectively and efficiently transfer between tasks. Currently, most transfer frameworks require substantial human intervention in at least one of the previous three steps. This discussion paper aims at: (1) positioning various knowledge re-use algorithms as forms of transfer, and (2) arguing the validity and possibility of autonomous transfer by detailing potential solutions to the above three steps.

AB - Reinforcement learning applications are hampered by the tabula rasa approach taken by existing techniques. Transfer for reinforcement learning tackles this problem by enabling the reuse of previously learned behaviours. To be fully autonomous a transfer agent has to: (1) automatically choose a relevant source task(s) for a given target, (2) learn about the relation between the tasks, and (3) effectively and efficiently transfer between tasks. Currently, most transfer frameworks require substantial human intervention in at least one of the previous three steps. This discussion paper aims at: (1) positioning various knowledge re-use algorithms as forms of transfer, and (2) arguing the validity and possibility of autonomous transfer by detailing potential solutions to the above three steps.

U2 - 10.1007/s13218-013-0286-8

DO - 10.1007/s13218-013-0286-8

M3 - Article

VL - 28

SP - 7

EP - 14

JO - Künstliche Intelligenz

JF - Künstliche Intelligenz

SN - 0933-1875

IS - 1

ER -