Presentation 2023-12-26
A study on selective reuse of local policies in transfer learning agents
Hiroya Hamada, Fumiaki Saitoh,
PDF Download Page PDF download Page Link
Abstract(in Japanese) (See Japanese page)
Abstract(in English) In recent years, reinforcement learning has gained attention for its application in acquiring AI behaviors. One challenge associated with reinforcement learning is the increase in the number of trials as tasks become more complex. To address this issue, transfer learning, which involves leveraging pre-learned knowledge to reduce the number of trials, has become a focus of interest. One method within transfer learning is εT-greedy, where applicable knowledge is randomly selected.However, a challenge arises in the selection of knowledge to transfer, as εT-greedy treats reusable and non-reusable knowledge equally, leading to an increase in unnecessary action selections. Therefore, this study proposes a method that assigns value to the knowledge to be selected and scales the selection probability accordingly.
Keyword(in Japanese) (See Japanese page)
Keyword(in English) Reinforcement learning / transfer learning / εT-greedy / knowledge interference
Paper # DE2023-29
Date of Issue 2023-12-19 (DE)

Conference Information
Committee DE / IPSJ-DBS
Conference Date 2023/12/26(1days)
Place (in Japanese) (See Japanese page)
Place (in English) Institute of Industrial Science, The University of Tokyo
Topics (in Japanese) (See Japanese page)
Topics (in English)
Chair Masashi Toyoda(Univ. of Tokyo) / 天笠 俊之(筑波大学)
Vice Chair Kosuke Takano(Kanagawa Inst. of Tech.) / Chiemi Watanabe(Ochanomizu Univ.)
Secretary Kosuke Takano(Komazawa Univ.) / Chiemi Watanabe(Univ. of Tsukuba) / (大阪大学)
Assistant Takahiro Komamizu(Nagoya Univ.) / Le Hieu Hanh(お茶の水女子大学)

Paper Information
Registration To Technical Committee on Data Engineering / Special Interest Group on Database System
Language JPN
Title (in Japanese) (See Japanese page)
Sub Title (in Japanese) (See Japanese page)
Title (in English) A study on selective reuse of local policies in transfer learning agents
Sub Title (in English)
Keyword(1) Reinforcement learning
Keyword(2) transfer learning
Keyword(3) εT-greedy
Keyword(4) knowledge interference
1st Author's Name Hiroya Hamada
1st Author's Affiliation Chiba Institute of Technology(CIT)
2nd Author's Name Fumiaki Saitoh
2nd Author's Affiliation Chiba Institute of Technology(CIT)
Date 2023-12-26
Paper # DE2023-29
Volume (vol) vol.123
Number (no) DE-327
Page pp.pp.7-11(DE),
#Pages 5
Date of Issue 2023-12-19 (DE)