Search bias, language bias and genetic programming

The use of bias with automated learning systems has become an important area of research. The use of bias with evolutionary techniques of learning has been shown to have advantages when complex structures are evolved. This is especially true when the semantics of the evolving population of structures is not explicitly represented or analysed during the evolution. This paper describes a framework which brings together two types of bias, namely search bias (the way new structures are created) and language bias (the form of possible structures that may be created). The resulting system extends genetic programming (GP) by allowing declarative bias with both the form of possible solutions that are created and the method by which they are transformed.