A Dynamic Algorithm Framework to Automatically Design a Multi-Objective Local Search

Metaheuristics are parametrized algorithms designed to solve complex optimization problems. Their parameters highly affect their performance and have to be set for each class of instance. Both offline and on-line approaches can be used to configure algorithms to be efficient.Offline approaches, also called automatic algorithm configuration (AAC), are able to handle many parameters but provide static algorithms adapted to the training instances only. On the other hand, on-line approaches provide adaptive algorithms whose parameters are modified during its execution but generally handle very few parameters. In this work, we propose a new model, called Dynamic Algorithm Framework in order to benefit from the advantages of both approaches.