Inspired by the idea of interacting intelligent agents of a multi-agent system, we introduce a multi-agent based optimization method applied to the quadratic assignment problem (MAOM-QAP). MAOM-QAP is composed of several agents (decision-maker agent, local search agents, crossover agents and perturbation agent) which are designed for the purpose of intensified and diversified search activities. With the help of a reinforcement learning mechanism, MAOM-QAP dynamically decides the most suitable agent to activate according to the state of search process. Under the coordination of the decision-maker agent, the other agents fulfill dedicated search tasks. The performance of the proposed approach is assessed on the set of well-known QAP benchmark instances, and compared with the most advanced QAP methods of the literature. The ideas proposed in this work are rather general and could be adapted to other optimization tasks. This work opens the way for designing new distributed intelligent systems for tackling other complex search problems.

Abstract. We propose a dedicated tabu search algorithm (TSX_WDP) for the winner determination problem (WDP) in combinatorial auctions. TSX_WDP integrates two complementary neighborhoods designed respectively for intensification and diversification. To escape deep local optima, TSX_WDP employs a backbone-based recombination operator to generate new starting points for tabu search and to displace the search into unexplored promising regions. The recombination operator operates on elite solutions previously found which are recorded in an global archive. The performance of our algorithm is assessed on a set of 500 well-known WDP benchmark instances. Comparisons with five state of the art algorithms demonstrate the effectiveness of our approach.

We propose a multi-agent based Distributed Hybrid algorithm for the Graph Coloring Problem (DH-GCP). DH-GCP applies a tabu search procedure with two different neighborhood structures for its intensification. To diversify the search into unexplored promising regions, two crossover operators and two types of perturbation moves are performed. All these search components are managed by a multi-agent model which uses reinforcement learning for decision making. The performance of the proposed algorithm is evaluated on well-known DIMACS benchmark instances.

This paper introduces a Multi-Agent based Optimization Method for Combinatorial Optimization Problems named MAOM-COP. In this method, a set of agents are cooperatively interacting to select the appropriate operators of metaheuristics using learning techniques. MAOM-COP is a flexible architecture, whose objective is to produce more generally applicable search methodologies. In this paper, the MAOM-COP explores genetic algorithm and local search metaheuristics. Using these metaheuristics, the decision-maker agent, the intensification agents and the diversification agents are seeking to improve the search. The diversification agents can be divided into the perturbation agent and the crossover agents. The decision-maker agent decides dynamically which agent to activate between intensification agents and crossover agents within reinforcement learning. If the intensification agents are activated, they apply local search algorithms. During their searches, they can exchange information, as they can trigger the perturbation agent. If the crossover agents are activated, they perform recombination operations. We applied the MAOM-COP to the following problems: quadratic assignment, graph coloring, winner determination and multidimensional knapsack. MAOMCOP shows competitive performances compared with the approaches of the literature.