Idealized Dynamic Population Sizing for Uniformly Scaled Problems

TL;DR

This paper proposes an idealized dynamic population sizing method for additive decomposable problems, aiming to approach the theoretical lower bounds of performance for self-adjusting algorithms.

Contribution

It introduces a population sizing strategy based on existing theory, closely approximating optimal fixed population performance for uniform scale problems.

Findings

The strategy performs near the theoretical lower bound.

It outperforms traditional fixed population methods.

Provides insights into self-adjusting population dynamics.

Abstract

This paper explores an idealized dynamic population sizing strategy for solving additive decomposable problems of uniform scale. The method is designed on top of the foundations of existing population sizing theory for this class of problems, and is carefully compared with an optimal fixed population sized genetic algorithm. The resulting strategy should be close to a lower bound in terms of what can be achieved, performance-wise, by self-adjusting population sizing algorithms for this class of problems.