Virtual Savant: learning for optimization

Renzo Massobrio, Sergio Nesmachnow, Bernabé Dorronsoro

Keywords: machine learning, combinatorial optimization, parallel computing, virtual savant

TL;DR: This article presents Virtual Savant, which derives knowledge from previously-solved optimization problem instances to solve unseen problem instances taking advantage of parallel computing.

Abstract: This article describes Virtual Savant, a novel paradigm that applies machine learning to derive knowledge from previously-solved optimization problem instances in order to solve new ones in a massively-parallel fashion. Applications of Virtual Savant to two classic combinatorial optimization problems and to one real-world problem are presented and experimental results are discussed.