Distributed Asynchronous Algorithm for Cross-Entropy-Based Combinatorial Optimization

Combinatorial optimization algorithms are used in many and diverse applications; for instance, in the planning, management, and operation of manufacturing and logistic systems and communication networks. For scalability and dependability reasons, distributed and asynchronous implementations of these optimization algorithms have obvious advantages over centralized implementations. Several such algorithms have been proposed in the literature. Some are inspired by what is known as ’swarm intelligence,’ e.g., ant-based optimization algorithms. Others are based on the method of cross-entropy. In this paper we present a generic distributed and asynchronous cross-entropybased algorithm for combinatorial optimization. The main advantages over previous algorithms include ease and flexibility of implementation, low overhead, robustness, and speed of convergence. As a result, the algorithm promises a wider applicability and significant efficiency gains compared to its predecessors, particularly for large-sized and real-time problems (such as those arising in distributed network management). Preliminary comparisons (based on empirical results using standard combinatorial problems) show that the proposed algorithm compares favourably with other existing distributed algorithms with respect to overhead and speed of convergence. Owing to its robustness and convergence properties, the proposed algorithm may be suited for real-time and dynamic applications.