Number of the records: 1
Similarity-based transfer learning of decision policies
- 1.
SYSNO ASEP 0534000 Document Type C - Proceedings Paper (int. conf.) R&D Document Type Conference Paper Title Similarity-based transfer learning of decision policies Author(s) Zugarová, Eliška (UTIA-B)
Guy, Tatiana Valentine (UTIA-B) RID, ORCIDNumber of authors 2 Source Title Proceedings of the IEEE INTERNATIONAL CONFERENCE ON SYSTEMS, MAN, AND CYBERNETICS 2020. - Piscataway : IEEE, 2020 - ISSN 1062-922X - ISBN 978-1-7281-8527-9 Pages s. 37-44 Number of pages 8 s. Publication form Online - E Action IEEE International Conference on Systems, Man and Cybernetics 2020 Event date 11.10.2020 - 14.10.2020 VEvent location Toronto Country CA - Canada Event type WRD Language eng - English Country US - United States Keywords probabilistic model ; fully probabilistic design ; transfer learning ; closed-loop behavior ; Bayesian estimation ; sequential decision making Subject RIV BB - Applied Statistics, Operational Research OECD category Statistics and probability R&D Projects LTC18075 GA MŠMT - Ministry of Education, Youth and Sports (MEYS) Institutional support UTIA-B - RVO:67985556 EID SCOPUS 85098853951 DOI 10.1109/SMC42975.2020.9283093 Annotation We consider a problem of learning decision policy from past experience available. Using the Fully Probabilistic Design (FPD) formalism, we propose a new general approach for finding a stochastic policy from the past data. The proposedapproach assigns degree of similarity to all of the past closed-loop behaviors. The degree of similarity expresses how close the current decision making task is to a past task. Then it is used by Bayesian estimation to learn an approximate optimal policy, which comprises the best past experience. The approach learns decision policy directly from the data without interacting with any supervisor/expert or using any reinforcement signal. The past experience may consider a decision objective different than the current one. Moreover the past decision policy need not to be optimal with respect to the past objective. We demonstrate our approach on simulated examples and show that the learned policy achieves better performance than optimal FPD policy whenever a mismodeling is present. Workplace Institute of Information Theory and Automation Contact Markéta Votavová, votavova@utia.cas.cz, Tel.: 266 052 201. Year of Publishing 2021
Number of the records: 1