This thesis investigates the following question: Can supervised learning techniques be successfully used for finding better solutions to multistage stochastic programs? A similar question had already been posed in the context of reinforcement learning, and had led to algorithmic and conceptual advances in the field of approximate value function methods over the years. This thesis identifies several ways to exploit the combination "multistage stochastic programming/supervised learning" for sequential decision making under uncertainty. Multistage stochastic programming is essentially the extension of stochastic programming to several recourse stages. After an introduction to multistage stochastic programming and a summary of existing approximation approaches based on scenario trees, this thesis mainly focusses on the use of supervised learning for building decision policies from scenario-tree approximations. Two ways of exploiting learned policies in the context of the practical issues posed by the multistage stochastic programming framework are explored: the fast evaluation of performance guarantees for a given approximation, and the selection of good scenario trees. The computational efficiency of the approach allows novel investigations relative to the construction of scenario trees, from which novel insights, solution approaches and algorithms are derived. For instance, we generate and select scenario trees with random branching structures for problems over large planning horizons. Our experiments on the empirical performances of learned policies, compared to golden-standard policies, suggest that the combination of stochastic programming and machine learning techniques could also constitute a method per se for sequential decision making under uncertainty, inasmuch as learned policies are simple to use, and come with performance guarantees that can actually be quite good. Finally, limitations of approaches that build an explicit model to represent an optimal solution mapping are studied in a simple parametric programming setting, and various insights regarding this issue are obtained.

Une décision perceptive est un processus délibératif consistant à choisir une proposition catégorielle ou un plan d'action parmi plusieurs alternatives sur la base d'information sensorielle. Les modèles de prise décision font l'hypothèse que l'information sensorielle est accumulée au cours du temps jusqu'à un seuil décisionnel. Ces modèles ont récemment reçu un support empirique important grâce à la découverte de neurones accumulateurs dans le cerveau de singes. Toutefois, l'étude neurophysiologique de ces système d'accumulation chez l'homme est rare. Ce travail de thèse vise à mieux comprendre les mécanismes neuronaux de prise de décision chez l'homme dans des contextes de la vie réelle, beaucoup plus complexes que ceux utilisés chez le singe. / A perceptual decision is a deliberative process that aims to choose a categorical proposition or course of action from a set of alternatives on the basis of available sensory information. Models of perceptual decision-making assume that sensory information is accumulated to some threshold level, whence the decision terminates in a choice. The recent discovery of neural correlates of these theoretical predictions in the non-human primate brain has reinforced their validity. However, neurophysiological studies of perceptual decision-making mechanisms in humans are relatively scarce. This work aims at enhancing our understanding of the computations and neurophysiology underpinning such mechanisms in humans, through the study of decision-making contexts more complex than those used in monkey research.