
Accession Number : AD0287460
Title : ADAPTIVE DECISION PROCESSES
Corporate Author : MASSACHUSETTS INST OF TECH CAMBRIDGE RESEARCH LAB OF ELECTRONICS
Personal Author(s) : ROSENFELD,JACK LEE
Report Date : 27 SEP 1962
Pagination or Media Count : 1
Abstract : General adaptive processes are described. In these processes a measure of performance is increased as the experimenter gathers more information; the actions taken by the experimenter determine both the profit and the type of information gathered. In particular, the adaptive decision process is a twoperson, zerosum, m X n game with some unknown payoffs. This game is played repeatedly. The true values of the unknown payoffs are learned only during those plays of the game at which the unknown payoffs are received. The players are given a priori probability distributions for the values of the unknown payoffs. A measure of performance is defined for the players of adaptive decision processes. An optimum strategy for one player is derived for the case in which the opponent uses one mixed strategy, known to the player, repeatedly. Optimum minimax strategies for both players are derived for the case in which the players are given the same information about the unknown payoffs. An optimum strategy, from a restricted clas of strategies, is derived for one player when he is playing against nature, which is assumed to be an opponent whose strategy is unknown but is unfavorable to the player. (Author)
Descriptors : *GAME THEORY, *PROBABILITY, ADAPTIVE CONTROL SYSTEMS, INEQUALITIES, MATHEMATICAL PREDICTION, MATRICES(MATHEMATICS), OPERATIONS RESEARCH, STATISTICAL ANALYSIS, STATISTICAL DISTRIBUTIONS
Distribution Statement : APPROVED FOR PUBLIC RELEASE