Stochastic Control: from Gradient Methods and Dynamic Programming to Statistical Learning

Controle stochastique: des méthodes de gradients à la programmation dynamique en passant par l'IA

Gilles Pagès (1) , Olivier Pironneau (2)

1 LPSM (UMR_8001) - Laboratoire de Probabilités, Statistique et Modélisation
2 LJLL - Laboratoire Jacques-Louis Lions

In this article the authors wish to contribute to the evaluation of statistical learning for stochastic control. We will review the well known methods for stochastic control and compare their numerical performance to those of a neural network. This will be done on a simple but practical example arising for fishing quotas to preserve the biomass of fish.

Mots clés: Partial Differential Equations; Dynamic Programming; Stochastic control; Statistical Learning

Domaines: Analyse numérique [cs.NA]; Intelligence artificielle [cs.AI]; Automatique; Analyse numérique [math.NA]

Gilles Pagès, Olivier Pironneau. Stochastic Control: from Gradient Methods and Dynamic Programming to Statistical Learning. 2020. ⟨hal-02911078v2⟩