Filippo Neri and Lorenza Saitta
In this paper the results of a preliminary study on the effects of selection and ordering of examples in incremental machine learning are presented. We consider learning as a two agents communication process, in which the teacher supplies information to the learner according to a prePdefined protocol, in order to transfer to him the knowledge sufficient to solve a given problem. The aim of this work is twofold: on one hand, to find out conditions under which, for a given protocol, the knowledge transfer can be successful and, on the other, to investigate how to exploit the example selection and ordering to speed up learning. A simulated robotic task has been chosen for the experiments, which have been performed with the availability of the system JWHY. The results of the experiments show that selection and ordering of examples can speed up learning, in terms of number of training examples considered.