Helmut Bölcskei - Fundamental limits of deep neural network learning

This talk was part of the workshop “MAIA 2019: Multivariate Approximation and Interpolation with Applications” held at the ESI August 26 – 30, 2019.

Deep neural networks have become state-of-the-art technology for a wide range of practical machine learningtasks such as image classification, handwritten digit recognition, speech recognition, or game intelligence. Thistalk develops the fundamental limits of learning in deep neural networks by characterizing what is possible if noconstraints on the learning algorithm and the amount of training data are imposed.