CS Special Seminar: Markus Heinonen "Differential Equations and Deep Learning"
Differential Equations and Deep Learning
Anstract: Deep models have revolutionised machine learning with recent developments of next-generation AIs for language and image synthesis that seem to demonstrate striking, imaginative domain understanding. Rather surprisingly, many such deep models have been shown to converge to differential equation structures, common in other sciences, but overlooked in machine learning. This fundamental connection opens new avenues of theory and practise on how to approach both dynamics and learning, and a new domain of continuous-time machine learning has emerged. In this job talk I will cover my research efforts in pushing the boundary of dynamical systems and deep statistical learning. I will also present my perspectives towards future developments.
Bio: Markus Heinonen is a machine learning Academy Research Fellow at Aalto University, with PhD from University of Helsinki in 2013. He studies dynamical systems, deep learning, probabilistic uncertainty, Gaussian processes, and their interconnections.