Jeff Johnston, Ph.D.

Jeff Johnston is a postdoctoral research scientist with Stefano Fusi in the Zuckerman Institute at Columbia University. He completed his graduate work with David Freedman at the University of Chicago, where he received an NRSA predoctoral fellowship from the NIH and the best dissertation award in computational neuroscience. Jeff’s work focuses on how the neural code enables reliable and flexible behavior, using a combination of approaches including data analysis, computer simulations and mathematical theory. Recently, Jeff has shown that “abstract” representations, which support the ability to generalize particular behaviors from one context to another, emerge naturally in artificial neural networks trained to perform multiple tasks, suggesting a method that may be at work in the brain. Jeff has also developed a geometric theory that captures the tradeoff between the representational geometry that enables this generalization and one that enables flexible behavior.

Principal Investigator: Stefano Fusi

Fellow: Mehnoor Khan

Project
The brain is often viewed as a modular system, which is split into multiple anatomically and functionally distinct brain regions. For example, in the primate visual system, certain brain regions are specialized for the representation of complex objects, while others are specialized to represent visual motion. Here, this project seeks to understand functional modularity in the brain by investigating when it emerges in artificial neural networks, as well as through the application of a previously developed mathematical theory. The researchers hypothesize that a system required to perform multiple distinct task-sets will split into modules at critical points predicted by the theory. The SURF fellow will help develop analytical tools for characterizing modularity and work from an established code base to test this hypothesis. The fellow will learn to train feedforward networks, apply common statistical models to data, and gain a background in related experimental and theoretical work on module formation.