Learning the “next” dimension

Abstract

In this paper we develop a novel search framework for optimization of functions over continuous domains based upon the building block hypothesis. We test one particular heuristic defined in terms of our framework (i.e. assumption 2, section 2) on a number of test functions and it exhibits promising performance. Since our heuristic is deterministic it is relatively easy to design a test function for which it fails. However, the search framework is general enough to define various other heuristics. Moreover, experience from search methods developed in the field of AI can be easily tailored in our search framework, since it basically represents a search in a tree structure. An important question to be addressed in future research is how to exploit the problem structure in order to define appropriate heuristics in the proposed search framework.

Another possible line for future research could be the utilisation of our search framework as a decision support tool that would interactively assist in the global optimization process.