Develop and reverse engineer continual learning algorithms

We can read a sentence and immediately incorporate it into our world view. This is really stunning. Brains integrate new information in real-time while preserving existing knowledge. Building on foundational memory consolidation research, my group integrates longitudinal neural recording data and new methods in machine learning to study various continual learning algorithms. This research will produce continually learning artificial systems and a better understanding of how biological systems learn over a lifetime.

Compartmentalized Computation : Modules in the Brain

The brain is composed of interconnected regions that perform specialized functions. Strokes and other brain traumas in addition to fMRI studies in humans and widefield imaging studies in mice provide an increasingly detailed map of the distinct functions across brain areas. However, there is a weak theoretical framework for what it means when we say two brain regions ‘interact’. Additionally, it is unclear what advantages are provided for neural networks through compartmentalized processing. In order to develop hypotheses for how modular networks perform computations, we train artificial neural networks with neural data informed architectures and reverse engineer these systems.

Multitasking Humans and RNNs : Interactions Across Computations in a Network

To advance our understanding of the flexibility of biological networks, we study how computations for multiple tasks are implemented within a single network of neurons. We employ a set of cognitive tasks for artificial recurrent neural networks (RNNs) and human participants enrolled in a brain-computer interface clinical trial to perform. We compare RNN dynamics to those recorded from populations of cortical neurons, with single neuron resolution using Neuropixel probes in human participants performing the same set of tasks. Following this novel approach, we develop models of human cognition and behavior.

Reverse engineer information seeking agents

I’m interested in the idea that animal behavior is information seeking. Human saccades on a face are directed towards the eyes and mouth rather than the cheeks, likely because these features provide more information to the viewer. It has been observed that some individuals with autism saccade in a more evenly distributed pattern across a face, rather than prioritizing highly informative regions. In order to better understand the computations involved in prioritizing where to look next, my group trains artificial systems to seek information and compare these systems to foraging mice. We study these two systems in a feedback loop to better understand how dynamical systems implement information seeking behavior.