Neural network approach for learning of the world structure by cognitive agents

Abstract

In this work an original method for coping with agents' incomplete knowledge is introduced. This method called the algorithm for the messages generation is applied by the cognitive agents when the states of external objects can not be directly perceived. To approximate the current states of objects all agent's experience as temporal data base is taken into account. As a result of the algorithm the logic formulas with modal operators are generated. One of the steps of proposed algorithm is the classification of the observations. It is shown how neural network approach might be used in order to determined some tendencies to occurrence specific states of objects. © Springer-Verlag Berlin Heidelberg 2006.