Abstract

A method of combining learning algorithms is described that preserves attribute efficiency. It yields learning algorithms that require a number of examples that is polynomial in the number of relevant variables and logarithmic in the number of irrelevant ones. The algorithms are simple to implement and realizable on networks with a number of nodes linear in the total number of variables. They can be viewed as strict generalizations of Littlestone's Winnow algorithm, and, therefore, appropriate to domains having very large numbers of attributes, but where nonlinear hypotheses are sought. 1 Introduction One area of stark contrast between current machine learning practice and natural biological learning is that of data preparation. In machine learning it is often found that in order to ensure success at a new learning task considerable effort has to be put into creating the right set of variables, and into eliminating redundant ones if there are large numbers of these. In biological lear...

Citations