KIAM Main page Web Library  •  Publication Searh  Ðóññêèé 
Publication

KIAM Preprint ¹ 170, Moscow, 2018
Authors: Knyazyatov S.A., Malinetskii G. G.
The solution of the problem of bluff detection in the game «I-doubt-it» based on reinforcement learning
Abstract:
In this paper we consider the construction of an algorithm based on reinforcement learning for the problem of recognizing and using a bluff on the example of a card game «I-doubt-it». The constructed algorithm has the 'intellectual ability' to restructure its behavior strategy and to evaluate possible moves based on previous experience.This class of algorithms used to make decisions in rapidly changing environments. The method and results of comparing algorithms among themselves, the results of games of the best algorithms with a real opponent are obtained. The effect of 'overfitting' is detected, increasing the number of training batches, in some cases, does not improve, but worsens the quality of the algorithm.
Keywords:
reinforcement learning, mathematical modeling, Q-learning, SARSA(λ) method, bluff detection algorithm, bluff imitation, neural networks, high-speed decision making
Publication language: russian,  pages: 21
Research direction:
Mathematical modelling in actual problems of science and technics
Russian source text:
Export link to publication in format:   RIS    BibTeX
View statistics (updated once a day)
over the last 30 days — 12 (+11), total hit from 01.09.2019 — 495
About authors:
  • Knyazyatov Stanislav Alexandrovich,  ,  Ñèáèíòåê
  • Malinetskii Georgii Gennadyevich,  orcid.org/0000-0001-6041-1926KIAM RAS