Friday, March 23, 2018, 3pm
Location: Planetary Hall Room 212

Tyrus Berry

Department of Mathematical Sciences
George Mason University

Forcasting without a model and with an imperfect model

Abstract
Often, time series data that appears to be high-dimensional is actually low-dimensional. I will introduce the diffusion maps algorithm, which identifies the low-dimensional manifold on which the data set lives, and discuss a recent application which uses this approach to perform model-free probabilistic forecasting of time series. This approach is optimal in a certain sense, and errors go to zero as the amount of training data goes to infinity. Finally, I will describe how this model-free approach can be combined with an imperfect model in order to overcome model error.