In the previous chapter, we introduced the concept of a computer system that could learn
how to perform a task by presentation of example inputs and desired outputs. In section 2.1
we formalise the challenge of creating such systems as the Supervised Learning Problem, and
introduce one class of learning system, neural networks. Several authors [66, 114, 115] have
noted that significant performance improvements can be obtained by creating a group of
learning systems, and combining their outputs in some fashion. In section 2.2 we review
current techniques for combining the outputs of a set of given predictors. In section 2.3 we
review techniques specifically for learning the predictors that are to be combined.