Neural classification of asymptotic (in)dependence

Troy Wixson (University of Massachusetts, Amherst)

The study of extremes allows us to learn about the tail of the distribution.
This research is important when studying rare events that have outsized impacts like heatwaves, wildfires, and hurricanes. These problems are inherently multivariate in nature and capturing dependence in the tail is challenging. Classifying a data set as asymptotically dependent (ADep) or asymptotically independent (AInd) is a necessary early choice in the modeling of multivariate extremes. In this talk I will introduce the study of extremes and then perform a series of experiments to determine whether a finite sample has enough information for a neural network to reliably distinguish between these regimes in the bivariate case. These experiments lead to a new classification tool for practitioners which we call nnadic as it is a Neural Network for Asymptotic Dependence/Independence Classification. This tool accurately classifies over 95% of test datasets and is robust to a wide range of sample sizes.
These experiments highlight that ADep and AInd models differ in whether the dependence completely decays in the limit, irrespective of the path of that decay.

Bio: Troy Wixson joined the UMass Department of Mathematics and Statistics as a Visiting Assistant Professor after completing his PhD in statistics at Colorado State University this past spring.
Troy has research interests related to modeling tail dependence in multivariate extremes with environmental applications and Bayesian modeling of heterogeneous data to probabilistically identify genes associated with disease.

Refreshments at 4:15pm, talk at 4:30pm