Preference Elicitation within Framework of Fully Probabilistic Design of Decision Strategies

0519795 - ÚTIA 2020 RIV GB eng C - Conference Paper (international conference)
Kárný, Miroslav - Guy, Tatiana Valentine
Preference Elicitation within Framework of Fully Probabilistic Design of Decision Strategies.
IFAC-PapersOnLine. Volume 52, Issue 29 - Proceedings of the 13th IFAC Workshop on Adaptive and Learning Control Systems 2019. Amsterdam: Elsevier, 2019, s. 239-244. ISSN 2405-8963.
[IFAC Workshop on Adaptive and Learning Control Systems 2019 /13./. Winchester (GB), 04.12.2019-06.12.2019]
R&D Projects: GA MŠMT(CZ) LTC18075
Grant - others:EU-COST(XE) CA16228
Institutional support: RVO:67985556
Keywords : dynamic decision making * Kullback Leibler Divergence * decision strategy * fully probabilistic design * preference elicitation
OECD category: Statistics and probability
http://library.utia.cas.cz/separaty/2019/AS/karny-0519795.pdf

The paper proposes the preference-elicitation support within the framework of fully probabilistic design (FPD) of decision strategies. Agent employing FPD uses probability densities to model the
closed-loop behaviour, i.e. a collection of all observed, opted and considered random variables. Opted actions are generated by a randomised strategy. The optimal decision strategy minimises KullbackLeibler divergence of the closed-loop model to its ideal counterpart describing the agent’s preferences. Thus, selecting the ideal closed-loop model comprises preference elicitation.
The paper provides a general choice of the best ideal closed-loop model reflecting agent’s preferences. The foreseen application potential of such a preference elicitation is high as FPD is a non-trivial dense extension of Bayesian decision making that dominates prescriptive decision theories. The general solution is illustrated on the regulation task with a linear Gaussian model describing the agent’s environment.
Permanent Link: http://hdl.handle.net/11104/0304785