Algorithmic configuration by learning and optimization

Gabriele Iommazzo , Claudia d'Ambrosio (1) , Antonio Frangioni (2) , Leo Liberti (1)

1 LIX - Laboratoire d'informatique de l'École polytechnique [Palaiseau]
2 Department of Computer Science [Pisa]

We propose a methodology, based on machine learning and optimization, for selecting a solver configuration for a given instance. First, we employ a set of solved instances and configurations in order to learn a performance function of the solver. Secondly, we solve a mixed-integer nonlinear program in order to find the best algorithmic configuration based on the performance function.

Domaines: Recherche opérationnelle [math.OC]; Apprentissage [cs.LG]

Gabriele Iommazzo, Claudia d'Ambrosio, Antonio Frangioni, Leo Liberti. Algorithmic configuration by learning and optimization. Cologne-Twente Workshop on Graphs and Combinatorial Optimization, Jul 2019, Twente, Netherlands. ⟨hal-02350282⟩