Title :
Multi-agent learning via implicit opponent modeling
Author :
Peterson, Teri S.
fDate :
6/24/1905 12:00:00 AM
Abstract :
We present a learning algorithm for two player stochastic games. The algorithm generates optimal deterministic finite automata (DFA) strategies against opponents who can be modeled by probabilistic action automata. The algorithm generates dynamic history trees based on statistical tests to eliminate state aliasing. Experiments are conducted in an iterated prisoner´s dilemma environment
Keywords :
deterministic automata; finite automata; learning (artificial intelligence); multi-agent systems; stochastic games; dynamic history trees; implicit opponent modeling; iterated prisoner´s dilemma environment; learning algorithm; multi-agent learning; optimal deterministic finite automata; probabilistic action automata; statistical tests; stochastic games; Computational modeling; Computer science; Doped fiber amplifiers; Game theory; Heuristic algorithms; History; Learning automata; Nash equilibrium; Stochastic processes; Testing;
Conference_Titel :
Evolutionary Computation, 2002. CEC '02. Proceedings of the 2002 Congress on
Conference_Location :
Honolulu, HI
Print_ISBN :
0-7803-7282-4
DOI :
10.1109/CEC.2002.1004470
