Algorithm Portfolios for Noisy Optimization: Compare Solvers Early

Marie-Liesse Cauwet (1, 2) , Jialin Liu (1, 2) , Olivier Teytaud (1, 2)

1 TAO - Machine Learning and Optimisation
2 LRI - Laboratoire de Recherche en Informatique

Noisy optimization is the optimization of objective functions corrupted by noise. A portfolio of algorithms is a set of algorithms equipped with an algorithm selection tool for distributing the compu- tational power among them. We study portfolios of noisy optimization solvers, show that different settings lead to dramatically different perfor- mances, obtain mathematically proved adaptivity by an ad hoc selection algorithm dedicated to noisy optimization. A somehow surprising result is that it is better to compare solvers with some lag; i.e., recommend the current recommendation of the best solver, selected from a comparison based on their recommendations earlier in the run.

Domaines: Optimisation et contrôle [math.OC]

Marie-Liesse Cauwet, Jialin Liu, Olivier Teytaud. Algorithm Portfolios for Noisy Optimization: Compare Solvers Early. Learning and Intelligent Optimization Conference, Feb 2014, Florida, United States. ⟨hal-00926638⟩