Imprecision and uncertainty in machine learning: a partial overview

 Sébastien Destercke (CNRS Sciences Informatiques/Heuristique et Diagnostic des Systèmes Complexes) 

In this talk, I will browse the different ways in which uncertainty and imprecision can be integrated to machine learning pipelines, focusing on the supervised setting. I will focus in particular on two aspects that are how to produce robust predictions in the form of sets, and how to integrate uncertain data in the learning framework. I will use examples to illustrate the various concepts addressed in the talk, and will finish by considering specific learning tasks in which integrating uncertainties may actually be an opportunity rather than a burden.