No abstract

This paper describes evolutionary programming, a systematic multi-agent stochastic search technique, used to generate recurrent perceptrons (nonlinear IIR filters). A hybrid optimization scheme is proposed that embeds a single-agent stochastic search technique, the method of Solis and Wets, into the evolutionary programming paradigm. The proposed hybrid optimization approach is further augmented by "blending" randomly selected parent vectors to create additional offspring. The first part of this work investigates the performance of the suggested hybrid Btochastic search method. After demonstration on the Bohachevsky and Rosenbrock response surfaces, the hybrid stochastic optimization approach is applied in determining both the model order and the coefficients of recurrent perceptron time-series models. An information criterion is used to evaluate each recurrent perceptron structure as a candidate solution. It is speculated that the stochastic training method implemented in this study for training recurrent perceptrons can be used to train perceptron networks that have radically recurrent architectures.

Oc~PItI:T nf)CUMENTATION PAGEIOM o. to average I hour per response. including fte time for reviewiing instruchori, ssarclung am"tin date ScsOMS. gathering anid AD-A273 '134 ______ :ion of inlorvation. Send comments regarding tIs burden esbrmate or any other a•pect of wits collection of •normation including Services. Directorate tor Into.ma.on Operatons and Rewps. 1215 Jelferson Davis Highway, Suite 1204. Arlington, VA rk Reduction Prolect (0704-0188). Washington. DC 20503. REPORT DATE REPORT TYPE AND DATES COVERED

No abstract

We use case injected genetic algorithms to learn how to competently play computer strategy games that involve long range planning across complex dynamics. Imperfect knowledge presented to players requires them adapt their strategies in order to anticipate opponent moves. We focus on the problem of acquiring knowledge learned from human players, in particular we learn general routing information from a human player in the context of a strike force planning game. By incorporating case injection into a genetic algorithm, we show methods for incorporating general knowledge elicited from human players into future plans. In effect allowing the GA to take important strategic elements from human play and merging those elements into its own strategic thinking. Results show that with an appropriate representation, case injection is effective at biasing the genetic algorithm toward producing plans that contain important strategic elements used by human players.