Learning and Forgetting with Local Information of New Objects

Abstract

The performance of supervised learners depends on the presence of a relatively large labeled sample. This paper proposes an automatic ongoing learning system, which is able to incorporate new knowledge from the experience obtained when classifying new objects and correspondingly, to improve the efficiency of the system. We employ a stochastic rule for classifying and editing, along with a condensing algorithm based on local density to forget superfluous data (and control the sample size). The effectiveness of the algorithm is experimentally evaluated using a number of data sets taken from the UCI Machine Learning Database Repository.