CHAPTER 14 - Representation Generation in an Exploratory Learning System

Abstract

This chapter discusses some of the features of DIDO, a program that has been developed to investigate possible solutions to the problems that arise in exploratory learning. The basic task is to acquire the knowledge necessary to solve problems in a novel domain, which is made up of discrete entities that DIDO can observe. The majority of machine learning systems can only operate successfully with the assistance of some external agent to provide guidance in the learning task. Such guidance takes various forms. The supervised concept learning systems, which learn classification rules from sets of classified examples, require an external agent to define the set of classes, to supply a set of training examples, and to classify the training examples. In contrast, conceptual clustering systems, which construct their own classification schemes from a set of training examples, only require the external agent to supply the examples.