6.1 Example strategies for agent model development

6.1.1 Crossover

In genetic programming (Koza [11]) the principle operation for model development is crossover. This is where random points in two models are selected and the appropriate submodels are swapped. This is a highly effective operator which has been used in solving many real-life problems. Its advantages are its speed, the fact it is universal in the sense that sequences of crossover operations can be used to construct any model (from a suitably large initial population), it preserves the variation in the population as well as the average depth of the population. From a perspective of modelling learning however, it is inappropriate due to its arbitrary action, the globality of its search and its need for very large populations of models to act upon.

An agent model can be effectively defined as a data structure with four slots: the independent variables, the dependent variables, the relationship between them and the conditions of application. Any adaptation of an agent model entails changes in the values of these slots. In this section we discuss coherent model-adaptation strategies in relation to the generalization and specialization of the agent's special models.

Modelling Learning as Modelling - 23 FEB 98

[Next] [Previous] [Up] [Top] [Contents]

Generated with CERN WebMaker