Event
Ph.D. Dissertation Defense: Joyneel Misra
Wednesday, November 5, 2025
12:00 p.m.
BPS 2145
Sarah Pham
301 473 2449
spham124@umd.edu
ANNOUNCEMENT: Ph.D. Dissertation Defense
Name: Joyneel Misra
Committee:
Professor Luiz Pessoa, Chair/Advisor
Professor Jonathan Z Simon
Professor Behtash Babadi
Professor Daniel Butts, Deans Representative
Professor Nan Xu
Date/time: November 5, 2025, 12:00 pm
Location: BPS 2145
Title: Study of Brain Dynamics in Complex and Dynamic Experiments
Abstract: In this dissertation, I try to advance analytical frameworks for studying brain dynamics to address the complexity of contemporary fMRI based experiments. Modern neuroscience is shifting toward complex, dynamic paradigms that try to better mimic real-world scenarios. This poses analytical challenges, such as loss of well defined events/trials during the experiment as basic units of analysis. Furthermore, recent works suggest trial-to-trial variability is not noise but corresponds to meaningful behaviors. Thus conventional trial-based averaging, which discards variability as noise, may be ill-suited for such experiments. This thesis aims to address these challenges faced, while analyzing brain activity as it unfolds dynamically during these complex tasks.
First, we apply Switching Linear Dynamical Systems (SLDS) to model fMRI data from a continuous threat-of-shock experiment, to propose dynamic brain-states as a new unit of data analysis. The model parsed fMRI timeseries into discrete brain states, which were successful at explaining the regularities of the experiment. By separating the dynamics equation into intrinsic and extrinsic (stimulus-driven) components, we show that the intrinsic dynamics of these states evolve toward stable fixed-point attractors. We further quantify how external inputs may contribute towards steering the system’s trajectory and driving transitions between states. Additionally, we introduce a novel "region importance" measure to determine which brain areas are most influential in steering these collective dynamics.
Next, we addressed trial-to-trial variability, hypothesizing that a repeated stimulus event may evoke multiple distinct dynamic responses instead of a single response. Analyzing a dynamic threat-avoidance task with SLDS, we represented each trial as a latent state sequence and clustered them to find multiple "dynamic modes", each characterized by a distinct brain response. The findings challenge the one-to-one stimulus-response mapping assumption. These dynamic modes were functionally significant, predictive of physiological arousal (skin conductance) and based on differences of the brain's engagement with stimuli, as per a controllability analysis. The study further validated the SLDS framework for uncovering meaningful trial-by-trial differences and linking neural patterns to threat-processing modes.
Finally, we develop an interpretable, supervised recurrent neural network (RNN) based framework to model spatiotemporal dynamics during naturalistic movie-watching. Using saliency maps and lesion analyses, we bridge the network's system-level representations to individual brain region contributions, identifying critical areas for decoding over time. Furthermore, spatiotemporal patterns learned by the model are generalizable across individuals and rich enough to predict stable cognitive traits, such as fluid intelligence, from a participant's dynamic brain response.
