Title: Continuous Networks for Sequential Predictions

Abstract: Deep learning is playing a growing role in many areas of science and engineering for modeling time series. However, deep neural networks are known to be sensitive to various adversarial environments, and thus out of the box models are often not suitable for mission critical applications. Hence, robustness and trustworthiness are increasingly important aspects in the process of engineering new neural network architectures and models. In this talk, I am going to view neural networks for time series prediction through the lens of dynamical systems. First, I will discuss novel continuous-time recurrent neural networks that are more robust and accurate than other traditional recurrent units. I will show that leveraging classical numerical methods, such as the higher-order explicit midpoint time integrator, improves the predictive accuracy of continuous-time recurrent units as compared to using the simpler one-step forward Euler scheme. Then, I will discuss a connection between recurrent neural networks and stochastic differential equations, and extensions such as multiscale ordinary differential equations for learning long-term sequential dependencies.



Bio Sketch: Ben Erichson is an Assistant Professor for Data-driven Modeling and Artificial Intelligence in the Department of Mechanical Engineering and Materials Science at the University of Pittsburgh. Before joining U Pitt, he was a postdoctoral researcher in the Department of Statistics at UC Berkeley, where he worked with Michael Mahoney. He was also a postdoc at the Department of Applied Mathematics at the University of Washington (UW) working with Nathan Kutz and Steven Brunton. He earned his PhD in Statistics at the University of St Andrews. Ben's work is broadly interested at the intersection of deep learning, dynamical systems, and robustness. He is also interested in leveraging tools from randomized numerical linear algebra to build modern algorithms for data-intensive applications such as fluid flows and climate science.