Solving Partially Observable Stochastic Games with Public Observations

Kód výsledku v IS VaVaI

<a href="https://www.isvavai.cz/riv?ss=detail&h=RIV%2F68407700%3A21230%2F19%3A00329800" target="_blank" >RIV/68407700:21230/19:00329800 - isvavai.cz</a>

Výsledek na webu

<a href="http://dx.doi.org/10.1609/aaai.v33i01.33012029" target="_blank" >http://dx.doi.org/10.1609/aaai.v33i01.33012029</a>

DOI - Digital Object Identifier

<a href="http://dx.doi.org/10.1609/aaai.v33i01.33012029" target="_blank" >10.1609/aaai.v33i01.33012029</a>

Jazyk výsledku

angličtina

Název v původním jazyce

Solving Partially Observable Stochastic Games with Public Observations

Popis výsledku v původním jazyce

n many real-world problems, there is a dynamic interactionbetween competitive agents. Partially observable stochasticgames (POSGs) are among the most general formal mod-els that capture such dynamic scenarios. The model capturesstochastic events, partial information of players about the en-vironment, and the scenario does not have a fixed horizon.Solving POSGs in the most general setting is intractable.Therefore, the research has been focused on subclasses ofPOSGs that have a value of the game and admit designing(approximate) optimal algorithms. We propose such a sub-class for two-player zero-sum games with discounted-sumobjective function—POSGs withpublic observations(PO-POSGs)—where each player is able to reconstruct beliefs ofthe other player over the unobserved states. Our results in-clude: (1) theoretical analysis of PO-POSGs and their valuefunctions showing convexity (concavity) in beliefs of maxi-mizing (minimizing) player, (2) a novel algorithm for approx-imating the value of the game, and (3) a practical demon-stration of scalability of our algorithm. Experimental resultsshow that our algorithm can closely approximate the value ofnon-trivial games with hundreds of states

Název v anglickém jazyce

Solving Partially Observable Stochastic Games with Public Observations

Popis výsledku anglicky

n many real-world problems, there is a dynamic interactionbetween competitive agents. Partially observable stochasticgames (POSGs) are among the most general formal mod-els that capture such dynamic scenarios. The model capturesstochastic events, partial information of players about the en-vironment, and the scenario does not have a fixed horizon.Solving POSGs in the most general setting is intractable.Therefore, the research has been focused on subclasses ofPOSGs that have a value of the game and admit designing(approximate) optimal algorithms. We propose such a sub-class for two-player zero-sum games with discounted-sumobjective function—POSGs withpublic observations(PO-POSGs)—where each player is able to reconstruct beliefs ofthe other player over the unobserved states. Our results in-clude: (1) theoretical analysis of PO-POSGs and their valuefunctions showing convexity (concavity) in beliefs of maxi-mizing (minimizing) player, (2) a novel algorithm for approx-imating the value of the game, and (3) a practical demon-stration of scalability of our algorithm. Experimental resultsshow that our algorithm can closely approximate the value ofnon-trivial games with hundreds of states

Druh

D - Stať ve sborníku

CEP obor

—

OECD FORD obor

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

Projekt

<a href="/cs/project/EF16_019%2F0000765" target="_blank" >EF16_019/0000765: Výzkumné centrum informatiky</a><br>

Návaznosti

P - Projekt vyzkumu a vyvoje financovany z verejnych zdroju (s odkazem do CEP)

Rok uplatnění

2019

Kód důvěrnosti údajů

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

Název statě ve sborníku

Proceedings of the Thirty-Third AAAI Conference on Artificial Intelligence

ISBN

978-1-57735-809-1

ISSN

2159-5399

e-ISSN

—

Počet stran výsledku

8

Strana od-do

2029-2036

Název nakladatele

AAAI Press

Místo vydání

Menlo Park, California

Místo konání akce

Honolulu

Datum konání akce

27. 1. 2019

Typ akce podle státní příslušnosti

WRD - Celosvětová akce

Kód UT WoS článku

000485292602006