Ph.D. Dissertation Defense: Proloy Das

Wednesday, August 5, 2020
12:00 p.m.
Zoom (Audio- and Video-enabled) meeting link https://umd.zoom.us/j/95342138302
Emily Irwin
301 405 0680
eirwin@umd.edu

ANNOUNCEMENT:  Ph.D. Dissertation Defense
 

Name: Proloy Das

Advisory Committee:

Professor Behtash Babadi, Chair
Professor Jonathan Z. Simon
Professor Shihab Shamma
Professor Steve Marcus
Professor Tom Goldstein
Professor Daniel A. Butts, Dean’s Representative

Time/Date: Wednesday, Aug 5, 2020 at 12 noon

Location: Zoom (Audio- and Video-enabled) meeting link https://umd.zoom.us/j/95342138302

 
Title: Bayesian Modeling and Estimation Techniques for the Analysis of Neuro-Imaging Data
 
Abstract: 
 
Brain function is hallmarked by its adaptivity and robustness, arising from underlying neural activity that admits well-structured representations in the temporal, spatial, or spectral domains. While neuroimaging techniques such as Electroencephalography (EEG) and magnetoencephalography (MEG) can record rapid neural dynamics at high temporal resolutions, they face several signal processing challenges that hinder their full utilization in capturing these characteristics of neural activity. The objective of this dissertation is to devise statistical modeling and estimation methodologies that account for the dynamic and structured representations of neural activity and to demonstrate their utility in application to experimentally-recorded data. 

The first part of this dissertation concerns spectral analysis of neural data. In order to capture the non-stationarities involved in neural oscillations, we integrate multitaper spectral analysis and state-space modeling in a Bayesian estimation setting. We also present a multitaper spectral analysis method tailored for spike trains that captures the non-linearities involved in neuronal spiking. We apply our proposed algorithms to both EEG and spike recordings, which reveal significant gains in spectral resolution and noise reduction. 

In the second part, we investigate cortical encoding of speech as manifested in MEG responses. These responses are often modeled via a linear filter, referred to as the temporal response function (TRF). While the TRFs estimated from the sensor-level MEG data have been widely studied, their cortical origins are not fully understood. We define the new notion of Neuro-Current Response Functions (NCRFs) for simultaneously determining the TRFs and their cortical distribution. We develop an efficient algorithm for NCRF estimation and apply it to MEG data, which provides new insights into the cortical dynamics underlying speech processing. 

Finally, in the third part, we consider the inference of Granger causal (GC) influences in high-dimensional time series models with sparse coupling. We consider a canonical sparse bivariate autoregressive model and define a new statistic for inferring GC influences, which we refer to as the LASSO-based Granger Causal (LGC) statistic. We establish non-asymptotic guarantees for robust identification of GC influences via the LGC statistic. Applications to simulated and real data demonstrate the utility of the LGC statistic in robust GC identification.

Audience: Public  Graduate  Faculty 

remind we with google calendar

 

October 2020

SU MO TU WE TH FR SA
27 28 29 30 1 2 3
4 5 6 7 8 9 10
11 12 13 14 15 16 17
18 19 20 21 22 23 24
25 26 27 28 29 30 31
1 2 3 4 5 6 7
Submit an Event