Deep Out-of-the-distribution Uncertainty Quantification for Data

This 57-minute talk by Nicolas Vayatis from ENS Paris-Saclay addresses the challenge of prediction diversity in deep learning when applied to out-of-distribution scenarios. Explore a practical solution that introduces the maximum entropy principle for weight distribution combined with standard in-distribution data fitting. Learn about the numerical proof demonstrating the systematic relevance of this algorithm and how this strategy can be applied to make out-of-distribution predictions about the future of data scientists. The presentation was delivered at the Institut des Hautes Etudes Scientifiques (IHES) and is available on CARMIN.tv, a French video platform specializing in mathematical content with functionalities designed for the research community.