Stay Informed:
Baskin Engineering COVID-19 Information and Resources
Campus Roadmap to Recovery
Zoom Links: Zoom Help | Teaching with Zoom | Zoom Quick Guide

AM seminar: Least-squares regression principle component analysis- a supervised dimension reduction method

Speaker Name: 
Xin (Cindy) Yee
Speaker Title: 
Assistant Professor, Mechanical and Aerospace Engineering
Speaker Organization: 
University of Colorado, Colorado Springs
Start Time: 
Monday, October 5, 2020 - 4:00pm
End Time: 
Monday, October 5, 2020 - 5:00pm
Location: 
Via Zoom Presentation
Organizer: 
Assistant Professor Marcella Gomez

Abstract:

One of the major challenges of applying machine learning (ML) methods to scientific problems is the construction of a set of most relevant descriptors/inputs to the ML models. In this talk, I will discuss the implications of the curse of dimensionality. I will then provide an overview of some existing dimensionality reduction methods. Lastly, I will motivate the need for a supervised dimensionality reduction method and present our new supervised dimensionality reduction method called least-squares regression principal component analysis (LSR-PCA). LSR-PCA bears many similarities as the popular principal component analysis (PCA) where we solve an eigenvalue problem. In addition, it can be used in conjunction with PCA further reduce irrelevant components of the input data. I will compare and contrast LSR-PCA with some of the existing supervised dimensional reduction methods.

Bio:

Xin (Cindy) Yee is an assistant professor in the Mechanical and Aerospace engineering department at the University of Colorado, Colorado Springs. Cindy received her Ph.D. from Caltech in 2015, and her B.S. from MIT in 2009. Her research interests include real-space and linear-scaling methods in Kohn-Sham density functional theory (KS-DFT), numerical methods for eigen-solvers for large sparse matrices, dimensionality reduction, machine learning.

Zoom Link: https://ucsc.zoom.us/j/93776214443?pwd=ZHBaMnJ4YUxEcFpHL25aUzZQWlJlUT09

Event Type: 
Event