SCGB Boston-Area Postdoc Meeting Series

Date & Time


Location

Samberg Conference Center
50 Memorial Drive, Chang Building (E52)
Cambridge, MA 02142 United States

View Map
Invitation Only

Participation is by invitation only. All participants must register.

The Simons Collaboration on the Global Brain invites you to the first Boston-area SCGB Postdoc Meeting. The purpose of these meetings is to bring together postdocs interested in neural coding and dynamics to discuss ideas and data. We will have two postdocs presentations, followed by dinner, drinks, and lively discussion.

The meeting is co-organized by Ali Yousefi, a postdoc at Harvard, and Sourish Chakravarty, a postdoc at MIT, and will take place at the Samberg Conference Center (6th Floor, Dining Room C).

The speakers for the first meeting are:

Seth W. Egger

Postdoctoral Associate at McGovern Institute for Brain Research, MIT

Internal models of sensorimotor integration regulate cortical dynamics

Theoretical considerations and psychophysical studies of sensorimotor integration have characterized the computational principles of motor control in terms of three building blocks: a controller, a simulator, and a state estimator. Although this characterization has had a profound impact on our understanding of control during overt movements, it is not known whether the same framework can explain the dynamic control of internal states in the absence of movements. Recently, it was shown that the brain controls the timing of movements by adjusting a speed command driving cortical dynamics. We leveraged this finding in a novel task in which the speed command had to be computed based on a sequence of flashes. Recordings from the frontal cortex of monkeys provided evidence that the brain uses a control strategy comprised of simulated motor plans to anticipate each upcoming flash, and estimation to update the speed after each flash. These findings suggest that internally-generated cortical dynamics supporting cognitive functions may be regulated by the same computational principles as motor control.

SueYeon Chung

BCS Fellow in Computation, MIT

Classification and geometry of neural manifolds, and the application to deep networks

Object manifolds arise when a neural population responds to an ensemble of sensory signals associated with different physical features (e.g., orientation, pose, scale, location, and intensity) of the same perceptual object. Object recognition and discrimination require classifying the manifolds in a manner that is insensitive to variability within a manifold. How neuronal systems give rise to invariant object classification and recognition is a fundamental problem in brain theory as well as in machine learning. We studied the ability of a readout network to classify objects from their perceptual manifold representations. We developed a statistical mechanical theory for the linear classification of manifolds with arbitrary geometries. We show how special anchor points on the manifolds can be used to define novel geometrical measures of radius and dimension, which can explain the linear separability of manifolds of various geometries. Theoretical predictions are corroborated by numerical simulations using recently developed algorithms to compute maximum margin solutions for manifold dichotomies. Our theory and its extensions provide a powerful and rich framework for applying statistical mechanics of linear classification to data arising from perceptual neuronal responses as well as to artificial deep networks trained for object recognition tasks. We demonstrate results from applying our method to both neuronal networks and deep networks for visual object recognition tasks. Exciting future work lies ahead as manifold representations of the sensory world are ubiquitous in both biological and artificial neural systems. Questions for future work include: How do neural manifold representations reformat in biological sensory hierarchies? Could we characterize dynamical neural manifolds for complex sequential stimuli and behaviors? How do neural manifold representations evolve during learning? Can neural manifold separability be used as a design principle for artificial deep networks?

Dinner and beverages will be served. Please forward this to colleagues that you think will be interested.

*You may be eligible for Uber Pool reimbursement to this event. Please email kscobie@simonsfoundation.org for more information.

We look forward to seeing you there!

Advancing Research in Basic Science and MathematicsSubscribe to SCGB announcements and other foundation updates