Policy Learning via Fully Probabilistic Design

Kód výsledku v IS VaVaI

<a href="https://www.isvavai.cz/riv?ss=detail&h=RIV%2F67985556%3A_____%2F24%3A00604532" target="_blank" >RIV/67985556:_____/24:00604532 - isvavai.cz</a>

Výsledek na webu

—

DOI - Digital Object Identifier

—

Jazyk výsledku

angličtina

Název v původním jazyce

Policy Learning via Fully Probabilistic Design

Popis výsledku v původním jazyce

Applying formalism of fully probabilistic design, we propose a new general data driven approach for finding a stochastic policy from demonstrations. The approach infers a policy directly from data without interaction with the expert or using any reinforcement signal. The expert’s actions generally need not to be optimal. The proposed approach learns an optimal policy by minimising Kullback-Liebler divergence between probabilistic description of the actual agent-environment behaviour and the distribution describing the targeted behaviour of the optimised closed loop. We demonstrate our approach on simulated examples and show that the learned policy: i) converges to the optimised policy obtained by FPD. ii) achieves better performance than the optimal FPD policy whenever a mismodelling is present.

Název v anglickém jazyce

Policy Learning via Fully Probabilistic Design

Popis výsledku anglicky

Applying formalism of fully probabilistic design, we propose a new general data driven approach for finding a stochastic policy from demonstrations. The approach infers a policy directly from data without interaction with the expert or using any reinforcement signal. The expert’s actions generally need not to be optimal. The proposed approach learns an optimal policy by minimising Kullback-Liebler divergence between probabilistic description of the actual agent-environment behaviour and the distribution describing the targeted behaviour of the optimised closed loop. We demonstrate our approach on simulated examples and show that the learned policy: i) converges to the optimised policy obtained by FPD. ii) achieves better performance than the optimal FPD policy whenever a mismodelling is present.

Druh

O - Ostatní výsledky

CEP obor

—

OECD FORD obor

10201 - Computer sciences, information science, bioinformathics (hardware development to be 2.2, social aspect to be 5.8)

Projekt

—

Návaznosti

I - Institucionalni podpora na dlouhodoby koncepcni rozvoj vyzkumne organizace

Rok uplatnění

2024

Kód důvěrnosti údajů

S - Úplné a pravdivé údaje o projektu nepodléhají ochraně podle zvláštních právních předpisů