Thursday, November 10, 2022, 2:30 – 3:30pm

Gaussian mixture models are collections of continuous probability distributions used for clustering and density estimation with applications in a variety of fields in science and engineering. Model parameters for Gaussian mixture models are typically estimated from training data using the iterative expectation-maximization (EM) algorithm, which requires knowing the number of Gaussian components a priori. In this study we propose an approach using numerical algebraic geometry to identify the optimal number of Gaussian components in a Gaussian mixture model. The proposed approach transforms a Gaussian mixture model into equivalent polynomial regression splines and uses homotopy continuation methods to find the model, or, equivalently, the number of components that is most compatible with the training data.

Zoom Participation. See announcement.

Event Type: Seminars
Room Number: Virtual Presentation - ET
Building: Remote Access - Zoom
Speaker's Name: ELIZABETH GROSS
Speaker Website: math.hawaii.edu…
Speaker's Professional Title: Associate Professor, Department of Mathematics, University of Hawaii at Mānoa
Talk Title: Model Selection for Gaussian Mixtures with Numerical Algebraic Geometry
For More Information: anezhad@andrew.cmu.edu
Affiliations: Computer Science Department (CSD), Tepper School of Business
Organization(s): Department of Mathematics