Few-Shots Parallel Algorithm Portfolio Construction via Co-Evolution

Generalization, i.e., the ability of solving problem instances that are not available during system design and development phase, is a critical goal for intelligent systems. A typical way to achieve good generalization learn model from vast data. In context heuristic search, such paradigm could be implemented as configuring parameters parallel algorithm portfolio (PAP) based on set “training” instances, which often referred PAP construction. However, compared traditional machine learning, construction suffers lack training obtained PAPs may fail generalize well. This article proposes novel competitive co-evolution scheme, named parameterized search (CEPS), remedy this challenge. By co-evolving configuration population an instance population, CEPS capable obtaining generalizable with few instances. The advantage in improving analytically shown article. Two concrete algorithms, namely, CEPS-TSP CEPS-VRPSPDTW, presented traveling salesman (TSP) vehicle routing simultaneous pickup-delivery time windows (VRPSPDTW), respectively. experimental results show has led better generalization, even managed find new best-known solutions some