GT MAP Seminar: Prof. Chethan Pandarinath (GT BME)

This is a part of the GT MAP activities on Control.  GT MAP is a place for research discussion and collaboration. We welcome participation of any researcher interested in discussing his/her project and exchange ideas with Mathematicians.

There will be light refreshments through out the event. This seminar will be held in Skiles 006 and refreshments at Skiles Atrium.

A couple of members of Prof. Pandarinath's group will present their research

3:00 PM - 3:45PM Prof.  Pandarinath will give a talk on "Unsupervised discovery of ensemble dynamics in the brain using deep learning techniques."

3:45PM -- 4:00PM Break with Discussions

4:00PM - 4:25PM Second talk given by Lahiru Wimalasena (Graduate student, Wallace H. Coulter Department of Biomedical Engineering, Emory University and Georgia Tech).

4:25PM - 5PM Discussion of open problems stemming from the presentations.

 

Title:  Unsupervised discovery of ensemble dynamics in the brain using deep learning techniques

Abstract: 

Since its inception, neuroscience has largely focused on the neuron as the functional unit of the nervous system. However, recent evidence demonstrates that populations of neurons within a brain area collectively show emergent functional properties ("dynamics"), properties that are not apparent at the level of individual neurons. These emergent dynamics likely serve as the brain’s fundamental computational mechanism. This shift compels neuroscientists to characterize emergent properties – that is, interactions between neurons – to understand computation in brain networks. Yet this introduces a daunting challenge – with millions of neurons in any given brain area, characterizing interactions within an area, and further, between brain areas, rapidly becomes intractable.

I will demonstrate a novel unsupervised tool, Latent Factor Analysis via Dynamical Systems ("LFADS"), that can accurately and succinctly capture the emergent dynamics of large neural populations from limited sampling. LFADS is based around deep learning architectures (variational sequential auto-encoders), and builds a model of an observed neural population's dynamics using a nonlinear dynamical system (a recurrent neural network).

When applied to neuronal ensemble recordings (~200 neurons) from macaque primary motor cortex (M1), we find that modeling population dynamics yields accurate estimates of the state of M1, as well as accurate predictions of the animal's motor behavior, on millisecond timescales. I will also demonstrate how our approach allows us to infer perturbations to the dynamical system (i.e., unobserved inputs to the neural population), and further allows us to leverage population recordings across long timescales (months) to build more accurate models of M1's dynamics.

This approach demonstrates the power of deep learning tools to model nonlinear dynamical systems and infer accurate estimates of the states of large biological networks. In addition, we will discuss future directions, where we aim to pry open the "black box" of the trained recurrent neural networks, in order to understand the computations being performed by the modeled neural populations.

pre-print available: lfads.github.io

 

Event Details

Date/Time:

  • Friday, March 30, 2018 - 3:00pm to 5:00pm

Location:
Skiles 006 Building

Fee(s):
_for_math_site_