Math Oracles: A New Way of Designing Efficient Self-Adaptive Algorithms

Abstract

In this paper we present a new general methodology to develop self-adaptive methods at a low computational cost. Instead of going purely ad-hoc we de ne several simple steps to include theoretical models as additional information in our algorithm. Our idea is to incorporate the predictive information (future behavior) provided by well-known mathematical models or other prediction systems (the oracle) to build enhanced methods. We show the main steps which should be considered to include this new kind of information into any algorithm. In addition, we actually test the idea on a speci c algorithm, a genetic algorithm (GA). Experiments show that our proposal is able to obtain similar, or even better results when it is compared to the traditional algorithm. We also show the bene ts in terms of saving time and a lower complexity of parameter settings.