(Colloquia from 3:30 - 4:30 p.m. in SH 140 with reception to follow in MLH 3)
Speaker: Sanvesh Srivastava, Ph.D., Assistant Professor and Director of Undergraduate Studies in Data Science and Statistics, Department of Statistics and Actuarial Science, College of Liberal Arts and Sciences, University of Iowa.
Abstract
Complex-valued tensor data arise in signal processing and neuroscience, where both magnitude and phase carry important information. I will present sparse separable factor analysis (SSFA), a latent factor model that learns structured dependence across tensor modes using low-rank
covariance models and sparse complex-valued loadings. Estimation uses a mode-wise parameter-expanded EM algorithm with complex soft-thresholding and a balancing step for scale identifiability. Simulations show substantial gains over vectorization-based methods for higher-order
data. I will illustrate the method using local field potential recordings, where SSFA captures dependence across brain region, frequency, and time and supports reconstruction of missing signals
Bio
Sanvesh Srivastava is currently an associate professor in the Department of Statistics and Actuarial Science at the University of Iowa. His research aims to develop flexible Bayesian methods and efficient computational algorithms for big data sets, tailored for both their complexity and size. Motivating examples include big data in genomics, medical imaging, and recommender systems. Simultaneously optimizing for the size and complexity is a challenge with current Bayesian methods. He is developing novel and computationally tractable Bayesian methods using principles from machine learning and optimal transportation.
Before coming to the University of Iowa, Sanvesh received his PhD in Statistics in August 2013 from Purdue University, where he also won I.W. Burr Award for "promise of contribution to the profession as evidenced by academic excellence in courses and exams, by the quality of research, and by excellence in teaching and consulting." After PhD, he spent two years at Duke University and Statistical and Applied Mathematical Sciences Institute (SAMSI) as a postdoctoral researcher. He has extensive experience in collaborating with scientists and teaching statistics to students from diverse areas and varied expertise.