Experience-based learning of semantic messages generation in resource-bounded environment
In this paper, an exploration of the results of agent's experience-based learning of semantic messages generation is presented. It is assumed that the agent is situated in some environment consisting of the atom objects. The agent observes the states of these objects and all the observations are stored in its private database. The agent is equipped with the communication language that makes it possible for an agent to generate modal formulas about the states of objects from external world. If the agent cannot observe the current state of a particular object then the algorithm for the choice of relevant semantic messages is used. This algorithm relates formulas to the internal agent's knowledge states and reflects the process of learning the world structure. An influence of agent's knowledge base changeability and the parameters of an algorithm on external messages generation are presented. Two alternative methods of consensus profile computing and two distance functions are used in this algorithm. A comparison of the results of an algorithm depending on its parameters is given.
Bibliogr. 17 poz., wykr.