Monday, November 28, 2022, 8am

Critical neurocognitive processes, such as maintaining circadian rhythms and performing natural activities, take place over minutes-to-days in chaotic, real-world environments. However, out of technical necessity, brain dynamics have been primarily studied on the scale of milliseconds-to-seconds while subjects are inside an artificial neural recording environment. Here we harness the rare opportunity to study brain dynamics during real-world behavior using intracranial electrode recordings in twenty humans for between 3-12 days of continuous recordings in each subject. During this time, subjects naturally interacted with friends and family, watched TV, slept, etc. while under simultaneous neural and video recordings. The functional networks that emerged possessed simple rules conserved over days that governed their individual dynamics, pairwise interactions, and relationships to circadian rhythm, arousal, and behavior. In contrast to single or paired network behavior, the mixture of all functional networks showed patterns of “punctuated equilibrium”: periods where networks would remain in stable states that corresponded to behavior and were interrupted by volatile transitions that were difficult to predict and displayed chaotic characteristics. Brain state statistics displayed characteristic power laws that are features of “self-organized criticality” – a characteristic of systems where complexity emerges from simple and stable building blocks. These results indicate that the complex and flexible brain dynamics that underpin real-world behavior are an emergent property of mixtures of individual, stable networks with simple dynamics.

In the second part of this project, we utilize the full week of data from each subject to learn interpretable nonlinear models that capture the overall network dynamics. We use a recurrent neural network and Koopman operator to learn a kernel that projects the original network feature space into an expanded non-linear feature set whose dynamics can be captured by standard discrete differential equations. Long-term dynamics of this expanded feature set are more accurately linked to natural behavior and can model behaviorally associated trajectories escaping a centroid attractor. This attractor naturally resembles the default mode network.

Thesis Committee:
Avniel Ghuman (Chair)
Robert Kass
Russell Schwartz
Tim Versynen
R. Mark Richardson (Massachusetts General Hospital)

Additional Information

In Person and Zoom Participation. See announcement.

Event Type: Thesis Proposals
Room Number: In Person and Virtual - ET
Building: Gates Hillman 8102 and Zoom
Speaker's Name: MAXWELL WANG
Speaker Website: sites.google.com…
Speaker's Professional Title: Ph.D. Student, Joint Ph.D. Program in Neural Computation and Machine Learning, Carnegie Mellon University
Talk Title: A week in the life of the human brain: stable states punctuated by chaotic-like transitions
For More Information: stidle@andrew.cmu.edu
Affiliations: Machine Learning Department (MLD)
Organization(s): Neuroscience Institute, SCS