Knowledge Representation in Machine Learning

Introduction Knowledge representation is a topic poorly discussed in machine learning. However, it is perhaps the fundamental consideration in the design of any learning system, because the representation used determines to a great degree what can and cannot be learned. This chapter presents a survey of different schemes used to represent learned knowledge in machine learning systems. Firstly, a brief description is given of the task of designing an adaptive system and of the central role which the method of representing initial and learned knowledge plays. Secondly we examine different representational schemes which have been used in learning systems, discussing the motivation for them, the way in which they are used and their limitations. The reader should note that we do not discuss representations appropriate for connectionist or genetic approaches to learning. This is not to dismiss them as less important, but merely to constrain the scope of this chapter to a tractable