Seminar Lecture by Prof. Christa Cuchiero
Prof. Christa Cuchiero from the Vienna University of Economics and Business will give a Seminar Lecture.
Note: The time has changed to 12:15 on Monday, 18 November 2019.
Title: Deep neural networks, generic universal interpolation and controlled differential equations
Abstract: "Deep neural networks can be viewed as discretizations of certain controlled ordinary differential equations. We make use of this perspective to link expressiveness of deep networks to the notion of controllability of dynamical systems. Using this connection, we study an expressiveness property that we call universal interpolation, and show that it is generic in a certain sense. We also show that universal interpolation holds for certain deep neural networks even if large numbers of parameters are left untrained, and instead chosen randomly. This lends theoretical support to the observation that training with random initialization can be successful even when most parameters are largely unchanged through the training.
On a related note we consider calibration of local stochastic volatility and yield curve prediction from finance and view these problems from an optimal control perspective. We parameterize the controls with neural networks and learn them directly from data without performing any kind of interpolation.
The talk is based on joint works with Wahid Khosrawi, Martin Larsson and Josef Teichmann."