Markus Heinonen: Differential equations and deep learning

Speaker: Markus Heinonen (Aalto University)

TitleDifferential equations and deep learning

Abstract: 

Differential equations describe the evolution of a system’s state, and are widely applied in natural sciences. Deep learning, on the other hand, is based on learning a sequential of the input towards a more complex state. Recently several groups have proposed to connect these, seeminly unconnected, branches of science together by re-imagining neural networks as executing a dynamical system, with pioneering works of Neural ODEs (Chen et al 2018) and deep Gaussian process flows (Hegde et al 2019).

In this talk I will describe how to model and learn high-dimensional ODEs and SDEs as Gaussian processes and neural networks, and how to employ them for deep learning. I will discuss the open problems when moving to dynamical system based machine learning.

Bio: 

PhD Markus Heinonen is an academy research fellow at Aalto University, Finland with focus on Bayesian deep learning, Gaussian processes, dynamical systems and their applications (https://users.aalto.fi/~heinom10)