Optimization for quantitative decisions: a versatile multi-tasker or an utopia?

Résumé En

The study of optimization algorithms started at the end of World War II and has since then experienced a constantly growing interest, fueled by needs in engineering, computational simulation and Machine Learning.In this talk, we look into the history of the new scientific objects that are the optimizers. The point-of-view is historical and philosophical rather than mathematical. It is explained how the conditions for the emergence of the optimizers as new scientific objects correspond to a state of sufficient knowledge decomposition. The versatility of optimizers is illustrated through examples. Although they are just tools, optimizers are a source a fascination because they guide through a space of representations in a process that resembles learning.But rationality is bounded and, in that sense, optimization problems are utopias: they are an over-simplification of decisions which are actually rooted in human relationships; they often cannot be solved because the computational capacities are limited.The presentation finishes with some contemporary challenges in optimization.