Optinformatics Across Heterogeneous Problem Domains and Solvers

Abstract

Besides the illustration of optinformatic algorithm design in evolutionary learning and optimization within a single problem domain, this chapter further introduce the specific algorithm development of optinformatics across heterogeneous problems or solvers. In particular, based on the paradigm of evolutionary search plus transfer learning in Sect. 3.2, the first optinformatic algorithm considers the knowledge transfer for enhanced vehicle or arc routing performance across problems domain. It intends to transfer the knowledge from vehicle routing to speed up the optimization of arc routing, and vice versa. The vehicle and arc routing problems used in Sect. 3.2 is again considered in this work to evaluate the performance of the optinformatic algorithm. Next, the second optinformatic method introduce the evolutionary knowledge learning and transfer across reinforcement learning agents in multi-agent system. Two types of neural network, i.e., feedforward and adaptive resonance theory (ART) neural network, are employed as the reinforcement learning agents, and the application of mine navigation and game of unreal tournament 2004 are used as the learning task to investigate the performance of the optinformatic method.