Ella Baker

Author: Ella Baker | Major: Biochemistry & Spanish | Semester: Fall 2024

I’m a physics and mathematics major, and I received a SURF grant for Spring/Fall 2024
to fund my research in Dr. Woodrow Shew’s lab in the physics department. My main
research interest is collective behavior of biological systems with many interacting
parts (like the brain, animal flocks, gene regulatory networks, etc.). I plan to pursue a
physics or computational neuroscience PhD.
I started working in the Shew Lab after taking Dr. Shew’s Thermal Physics class in
Spring 2022. My research focuses on the hypothesis that cerebral cortex of awake
animals operates close to a boundary between different dynamical regimes, a state
called “criticality.” At criticality, fluctuations have no characteristic spatial or temporal
scale and many aspects of information processing are optimized. These functional
advantages of criticality suggest that it may be an evolutionary endpoint for the brain,
and that neurological disease or dysfunction may coincide with deviation from
criticality. Progress on these ideas has been limited, however, by the absence of an
accepted way to measure deviation from criticality.
During the funding period of this SURF grant, I developed a partial solution to this
problem. Specifically, I first defined distance to criticality in terms of an
information-theoretic measure of distinguishability between models called KL
divergence. Next, I applied the renormalization group (RG)–a theoretical tool originally
developed to understand the long wavelength behavior of non-living systems like
magnets–to a fundamental class of time series models, autoregressive (AR) models.
The idea of RG is to track how a model changes under a change of scale; a model at
criticality flows into an RG fixed point, and different fixed points correspond to different
types (different universality classes) of critical dynamics. For AR models, there is a
nested hierarchy of RG fixed points, and the sets of AR models that flow into them form
extended basins of attraction. Knowing which AR models are at criticality, we can
explicitly calculate distance to criticality in the space of AR models.
This immediately gives rise to a pipeline for measuring distance to criticality from time
series data: first fit the data to an AR model, then calculate the AR model’s distance to
criticality. I applied this approach to spike recordings from visual cortex of freely-moving
mice, finding that cortex is closest to criticality at an intermediate level of arousal,
between the extremes of deep sleep and the hyper-aroused awake state. Also, in line
with a previous report, I found that sleep homeostatically restores nearness to criticality;
cortex is usually farther from criticality at the beginning of sleep than at the end.
During the Fall 2024 semester, I traveled to two conferences to present this work: the
Society for Neuroscience (SfN) conference and the Brain Criticality Conference, both in
Washington, DC. I had been to SfN the previous year also, so I wasn’t surprised by the
huge crowds (~10000 attendees). The Brain Criticality Conference, on the other hand,
was different from any of my previous conference experiences; it was held in a small
room and had roughly 100 attendees, most of whom knew each other. This spring, I’m
excited to present a poster at COSYNE (COmputational and SYstems NeurosciencE) and
finish preparing the work described in this blog post for publication