A Design Approach to Problem Difficulty
The last chapter motivated and expanded upon the notion of a building block to understand the raw material available for genetic search. This chapter builds on these ideas to understand the ways in which a GA might succeed or fail depending upon the type of problem it faces. Our approach is to design bounding adversarial problems that represent different dimensions of problem difficulty. Success in this endeavor will allow us to test different algorithms against a limited number of boundedly difficult problems in such a way that algorithm success against the particular problems ensures success against a large class of problems no harder than the test cases. This adversarial design method contrasts sharply with the common practices of using historical, randomly generated, or ad hoc test functions.
KeywordsBuilding Block Problem Difficulty Design Envelope Archetypal Model Trap Function
Unable to display preview. Download preview PDF.