Favourable Biasing of Function Sets Using Run Transferable Libraries

Abstract

This chapter describes the notion of Run TransferableLibraries(RTLs), libraries of functions which evolvefrom run to run. RTLs have much in common with standardprogramming libraries as they provide a suite offunctions that can not only be used across several runson a particular problem, but also to aid in the scalingof a system to more difficult instances of a problem.This is achieved by training a library on a relativelysimple instance of a problem before applying it to themore difficult one.The chapter examines the dynamics of the libraryinternals, and how functions compete for dominance ofthe library. We demonstrate that the libraries tend toconverge on a small number of functions, and identifymethods to test how well a library is likely to be ableto scale.