Anomalies and Commonalities
Michael Lindstrom (University of Texas Rio Grande Valley)
Abstract: This talk will cover two research projects. In the first, we seek to identify contextual anomalies among time series with possible missing data. By generalizing Kernel Density Estimation to Hilbert Spaces, we develop tools to identify anomalous time series, test them against competing methods on synthetic data, and then employ the tools to identify anomalous event records among airplane fleets. In the second project, we combine topic modelling through Nonnegative Matrix Factorization and regression on a continuous observation variable. Previous authors have covered the case of topic modelling for classification; here, we show the idea can be extended to regression, applying it to Rate My Professors reviews and predicting an instructors mean rating from the written comments. We identify interpretable groups of words (topics), such that their level of representation in a review has a quantifiable effect on the associated rating.
BiologyMathematicsPhysics
Audience: researchers in the topic
Series comments: https://oklahoma.zoom.us/j/99460473420
Organizers: | Rongchang Liu, Christian Parkinson*, Weinan Wang* |
*contact for this listing |