Dynamic hill climbing: Overcoming the limitations of optimization techniques

Deniz Yuret and Michael de la Maza (1993) ( PS ): Dynamic hill climbing: Overcoming the limitations of optimization techniques. In The Second Turkish Symposium on Artificial Intelligence and Neural Networks.

Abstract:

This paper describes a novel search algorithm, called dynamic hill climbing, that borrows ideas from genetic algorithms and hill climbing techniques. Unlike both genetic and hill climbing algorithms, dynamic hill climbing has the ability to dynamically change its coordinate frame during the course of an optimization. Furthermore, the algorithm moves from a coarse-grained search to a fine-grained search of the function space by changing its mutation rate and uses a diversity-based distance metric to ensure that it searches new regions of the space. Dynamic hill climbing is empirically compared to a traditional genetic algorithm using De Jong's well-known five function test suite and is shown to vastly surpass the performance of the genetic algorithm, often finding better solutions using only 1% as many function evaluations.