2019 Mar 15, Jyotishka Datta, University of Arkansas New directions in Bayesian sparse signal recovery
From Kathie Leck
Faculty, staff, and students are required to read, understand and comply with the PSU Copyright Policy and all applicable law regarding materials uploaded to Kaltura. Failure to comply with law and PSU policy may result in individual liability.This Policy can be found here.
Full Transcript File located below Media Player
New directions in Bayesian sparse signal recovery
Sparse signal recovery remains an important challenge in large scale data analysis and global-local (G-L) shrinkage priors have emerged as the current state-of-the art Bayesian method handling sparsity. In the first half of this talk, I will survey some of the recent theoretical and methodological advances in this area, focusing on theoretical optimality of G-L priors in the context of multiple testing for both continuous as well as quasi-sparse count data. In the second half, I will discuss a few unexplored aspects of their behavior, such as, validity as a non-convex regularization method, performance in presence of correlated errors or extension to discrete data structures including sparse compositional data. I will offer some insights into some of these problems and point out future directions.