Ph.D. Research Proposal: Srinivas Govinda Surampudi

Monday, November 17, 2025
10:00 a.m.
AVW (ECE) 2328
Sarah Pham
301 473 2449
spham124@umd.edu

ANNOUNCEMENT: Ph.D. Research Proposal Exam

 

Name: Srinivas Govinda Surampudi

 

Committee:

Professor NAME (Chair) Prof. Luiz Pessoa

Professor NAME Prof. Jonathan Z Simon

Professor NAME Prof. Behtash Babadi

 

Date/time: November 17, 2025  from 10:00am to 12:00pm

 

Location: AVW (ECE) 2328

 

Title: Bayesian inference reveals multiple plausible architectures in the mouse functional connectome

 

Abstract: 

Functional brain architecture is often simplified to a single, static map of distinct modules. This view, however, struggles to account for the brain's complex organization and often treats the inherent variability found across different analyses as methodological noise. We employ a principled Bayesian generative modeling framework to investigate the community structure of resting-state functional connectivity networks from the mouse brain. Our approach allows us to first formally test competing theories of network organization and then to characterize the full landscape of plausible solutions, moving beyond a single consensus map. Our analysis revealed that the principle of segregated modules (assortativity) was only weakly supported by the data. Instead, a hierarchical arrangement of communities of brain regions provided the most faithful explanation of the functional brain network. Critically, this model revealed that the brain's organization is not a single map but a landscape of multiple, coexisting configurations. We found this landscape is underpinned by a library of recurring, archetypal compositions of brain regions for each community. These archetypes revealed a spectrum of stability and flexibility---from highly stable, foundational units like the somatomotor system to flexible, integrative systems like the medial temporal pole. Our approach encourages the field to go beyond the "single-map" paradigm. We propose that viewing brain's functional architecture as a probabilistic ensemble of plausible arrangements of archetypal compositions offers a more rigorous and nuanced framework for understanding its complex and flexible nature.

Audience: Graduate  Faculty 

remind we with google calendar

 

October 2025

SU MO TU WE TH FR SA
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
Submit an Event