Event
Ph.D. Dissertation Defense: Srinivas Govinda Surampudi
Wednesday, June 17, 2026
1:00 p.m.
AVW 2460
Emily Irwin
301 405 0680
eirwin@umd.edu
ANNOUNCEMENT: Ph.D. Dissertation Defense
Name: Srinivas Govinda Surampudi
Committee:
Professor Luiz Pessoa, Chair/Advisor
Professor Jonathan Z Simon
Professor Behtash Babadi
Professor Nan Xu
Professor Joshua Singer, Dean's Representative
Professor Behtash Babadi
Professor Nan Xu
Professor Joshua Singer, Dean's Representative
Date/time:
June 17 2026, 1:00pm - 3:00pm
Location:
AVW 2460
Zoom link: https://umd.zoom.us/j/
Passcode: defense
Title:
A probabilistic ontology of functional brain organization: evidence from rodent resting-state and human affective-task connectomes
Abstract:
Understanding the principles that govern the macroscopic functional organization of the brain remains a central challenge in systems neuroscience. The field has primarily relied on descriptive clustering heuristics, such as modularity maximization, to extract a single, consensus network partition from functional magnetic resonance imaging (fMRI) data. While highly useful for charting foundational brain geography, this deterministic approach implicitly assumes a uni-modal network architecture, potentially obscuring the structural multiplicity inherent in complex neural systems. This dissertation addresses this limitation conceptually and empirically by applying a Bayesian generative modeling framework—specifically, non-parametric Stochastic Block Models (SBMs)—to shift the analytical focus from extracting a single optimal partition toward characterizing the full posterior landscape of organizational hypotheses.
This probabilistic framework is applied to two distinct functional neuroimaging datasets to investigate brain organization across different species and cognitive states. First, evaluating a high-resolution resting-state fMRI dataset from rodents, we demonstrate that a non-degree-corrected hierarchical architecture provides the most parsimonious explanation of the baseline functional connectome. Crucially, the posterior landscape reveals that this connectome is structurally multi-modal, supported by multiple, co-dominant organizational modes driven by the targeted flexibility of associative routing hubs. Second, we extend this framework to the human brain during states of affective imminence, utilizing dynamic threat-avoidance and reward-pursuit tasks. We find that the human connectome similarly favors a hierarchical, degree-homogeneous architecture and maintains a conserved repertoire of co-dominant organizational modes. Rather than wholesale reorganization of community boundaries, affective task demands systematically reweight the posterior occupancy of these existing modes along a shared salience-action-control axis.
Collectively, these empirical applications demonstrate that the static functional connectome is fundamentally underdetermined by a single consensus map. By framing community detection as a process of statistical inference, this dissertation advocates for a probabilistic ontology of large-scale brain organization, demonstrating that preserving the multiplicity of the posterior landscape provides a more accurate, comprehensive representation of the connectome's functional capacity.
