research

Sensorimotor integration and the geometry of neural population dynamics, studied in songbirds.

I work at the intersection of computational and systems neuroscience, motivated by neuroethology and the challenge of linking real-world behavior to neural activity.

birdsong auditory forebrain NCM · CMM song motor pathway HVC · RA hearing singing

The sounds songbirds have to interpret while listening are the same sounds they later produce and sequence during singing. One behavior gives two views of the same problem.

current work

neural dimension 1 neural dimension 2 category A correct category B decision boundary not expected expected dots mark equal steps in time

Each line is the dynamics of the auditory population on one trial, moving through a space where sounds from two categories separate.

01 · listening

Expectation organizes population dynamics

How does knowing what is coming change the way a sensory population responds? In songbirds categorizing natural song syllables, expectation reorganized the dynamics of the auditory forebrain. Neural population responses to sounds from the same category grew more alike, the opposite of what single neurons suggest, and the early motion of population activity on each trial predicted how quickly and accurately the bird chose. Expectation works as a form of control, setting the initial conditions from which a decision unfolds.

premotor region neural population motor region neural population drives feeds back how strongly the two regions interact silence song silence song time

Two neural populations recorded at the same time. Their interaction changes from moment to moment as the song unfolds.

02 · singing

Song is controlled by a coupled loop

A learned song has to be commanded ahead of time, because signals take time to travel from brain to voice. The standard picture is a chain of command, in which a premotor region drives a motor region. Using simultaneous recordings of both neural populations, I find a loop where each region's dynamics shape the other's, and the strength of that interaction rises and falls with the structure of the song. Treating the pair as a single coupled dynamical system suggests how the circuit keeps a fast sequence under control despite its own delays.

changes the output silent to the output prepare act

Schematic of one trial. The dot marks where preparation ends and action begins.

03 · framework

The same question, asked of both

A population can change in ways that alter what it sends downstream and in ways that do not. Separating the two shows how a brain can prepare without yet acting, in hearing and in singing alike.

where it is going

As a recipient of the NIH F99/K00 Predoctoral to Postdoctoral Transition Award, I will carry this work into a postdoc in the Brunton Lab at the University of Washington beginning in 2027. My overarching goal is to create testable models that connect circuit architecture, neural population coding, and flexible cognition, and that yield generalizable theoretical principles for large-scale neural activity.

how I got here

My research experience spans a range of disciplines.

papers

Publications and preprints are on Google Scholar.