Investigating Case Learning Techniques for Agents to Play the Card Game of Truco

Moral, Ruan C. B.; Paulus, Gustavo Bathu; Assunção, Joaquim; Silva, Luís A. L.

doi:10.1109/sbgames51465.2020.00024

Cited by 5 publications

(2 citation statements)

References 13 publications

Supporting

Mentioning

Contrasting

Order By: Relevance

“…However, any agent modeled with such structures has a problem not approached in any know work [Paulus et al 2019, Moral et al 2020, Vargas et al 2021, Rossato et al 2020; as the game rolls the personality of the player or the approach of the agent, can be learned. Such a task is easy for humans but not so for AI.…”

Section: Proposed Modelmentioning

confidence: 99%

“…A few works used Case-Based Reasoning (CBR) [Richter and Weber 2013] as an AI framework to solve problems in the game of Truco [Paulus et al 2019, Moral et al 2020, Vargas et al 2021, such approach relies on creating a database of cases for the game, latter exploring the data with different approaches. Specifically, [Paulus et al 2019] did a major comparison between shallow learning approaches, [Vargas et al 2021] built a database focusing on bluff and [Rossato et al 2020], without using CBR, built a lightweight Markov model using the natural force of all possible hands.…”

Section: Introductionmentioning

confidence: 99%

See 1 more Smart Citation

Towards a time-aware HMM for the Truco game

Assunção

Paulus²

2022

Anais Estendidos Do XXI Simpósio Brasileiro De Jogos E Entretenimento Digital (SBGames Estendido 2022)

View full text Add to dashboard Cite

Similar to Poker, the game of Truco has challenges for Artificial intelligence. Considering a large number of game states, a scenario characterized by partial visibility, stochastic behavior, and score susceptible to bluff; this game offers a good set of rules to test and improve AI techniques. In this article, we describe the creation of a Hidden Markov Model (HMM) agent using temporal control. The model has an embedded vector that adjusts its probabilities for further game actions, consequently, improving the model playing performance. The evaluation is given with over 210,000 matches, serving as empirical proof of the idea.

show abstract

Section: Proposed Modelmentioning

confidence: 99%