Learning Appropriate Contexts

Genetic Programming is extended so that the solutions being evolved do so in the context of  local domains within the total problem domain.  This produces a situation where different “species” of solution develop to exploit different “niches” of the problem – indicating exploitable solutions.  It is argued that for context to be fully learnable a further step of abstraction is necessary.  Such contexts abstracted from clusters of solution/model domains make sense of the problem of how to identify when it is the content of a model is wrong and when it is the context.  Some principles of learning to identify useful contexts are proposed.