Model generation generally requires two components: a model representation and a search algorithm to search through it. Evolutionary algorithms, as general-purpose search algorithms, may be used for the latter, even within learning methods that are not intrinsically evolutionary. They are most widely used in neural networks, to learn both the weights, and the architectures, of the networks. However, they have also been applied to learning decision trees, learning kernel functions for support vector machines (see Support Vector Machines), and evolving functional dependency networks. They are heavily used in ensemble methods, in which a number of models are built in parallel, then used jointly in prediction.
Was this article helpful?