Why operator-induced neighborhoods are a bad idea

Abstract

All of the solutions to a combinatorial optimization problem can be collected into a fitness landscape. Stochastic searches for the optimal solution are usually employed because the landscapes are so large. Studies have shown search operators that exploit the landscape structure are the most computationally efficient, which drives the research interest in discovering landscape structure. Some researchers claim this structure is induced by the search operator itself. In this paper we show any structural information obtained from operator-induced neighborhoods is virtually useless.

Keywords:

• Combinatorial optimization
• fitness landscape