| Name: | Description: | Size: | Format: | |
|---|---|---|---|---|
| 6.82 MB | Adobe PDF |
Authors
Advisor(s)
Abstract(s)
A combinação de princípios da teoria de jogo e metodologias de machine learning aplicados ao contexto de formular estratégias ótimas para jogos está a angariar interesse por parte de uma porção crescentemente significativa da comunidade científica, tornando-se o jogo do Poker num candidato de estudo popular devido à sua natureza de informação imperfeita. Avanços nesta área possuem vastas aplicações em cenários do mundo real, e a área de investigação de inteligência artificial demonstra que o interesse relativo a este objeto de estudo está longe de desaparecer, com investigadores do Facebook e Carnegie Mellon a apresentar, em 2019, o primeiro agente de jogo autónomo de Poker provado como ganhador num cenário com múltiplos jogadores, uma conquista relativamente à anterior especificação do estado da arte, que fora desenvolvida para jogos de apenas 2 jogadores. Este estudo pretende explorar as características de jogos estocásticos de informação imperfeita, recolhendo informação acerca dos avanços nas metodologias disponibilizados por parte de investigadores de forma a desenvolver um agente autónomo de jogo que se pretende inserir na classificação de "utility-maximizing decision-maker".
The combination of game theory principles and machine learning methodologies applied to encountering optimal strategies for games is garnering interest from an increasing large portion of the scientific community, with the game of Poker being a popular study subject due to its imperfect information nature. Advancements in this area have a wide array of applications in real-world scenarios, and the field of artificial intelligent studies show that the interest regarding this object of study is yet to fade, with researchers from Facebook and Carnegie Mellon presenting, in 2019, the world’s first autonomous Poker playing agent that is proven to be profitable while confronting multiple players at a time, an achievement in relation to the previous state of the art specification, which was developed for two player games only. This study intends to explore the characteristics of stochastic games of imperfect information, gathering information regarding the advancements in methodologies made available by researchers in order to ultimately develop an autonomous agent intended to adhere to the classification of a utility-maximizing decision-maker.
The combination of game theory principles and machine learning methodologies applied to encountering optimal strategies for games is garnering interest from an increasing large portion of the scientific community, with the game of Poker being a popular study subject due to its imperfect information nature. Advancements in this area have a wide array of applications in real-world scenarios, and the field of artificial intelligent studies show that the interest regarding this object of study is yet to fade, with researchers from Facebook and Carnegie Mellon presenting, in 2019, the world’s first autonomous Poker playing agent that is proven to be profitable while confronting multiple players at a time, an achievement in relation to the previous state of the art specification, which was developed for two player games only. This study intends to explore the characteristics of stochastic games of imperfect information, gathering information regarding the advancements in methodologies made available by researchers in order to ultimately develop an autonomous agent intended to adhere to the classification of a utility-maximizing decision-maker.
Description
Keywords
Machine Learning Game Theory Poker Strategy Games Probabilities Opponent Modeling