Revisiting the auto-regressive functions of the cross-entropy ant system

Abstract

The Cross-Entropy Ant System (CEAS) is an Ant Colony Optimization (ACO) system for distributed and online path management in telecommunication networks. Previous works on CEAS have enhanced the system by introducing new features. This paper takes a step back and revisits the auto-regressive functions at the core of the system. These functions are approximations of complicated transcendental functions stemming from the Cross-Entropy (CE) method for stochastic optimization, computationally intensive and therefore not suited for online and distributed operation. Using linear instead of hyperbolic approximations, new expressions are derived and shown to improve the adaptivity and robustness of the system, in particular on the occurrence of radical changes in the cost of the paths sampled by ants. © 2009 Springer-Verlag.