CAREER: RI: Uncovering principles of robust intelligence by disentangling the functional architecture of the brain

Multiple systems of the brain are coordinated to produce the intelligence that enables complex behaviors. While data collections tools have been steadily increasing the ability to record activity from across the entire brain, analyses of these data have primarily focused on dividing neurons by anatomical areas and analyzing neural activity within and across brain areas.

Recent observations suggest, however, that coordination of brain activity is related to, but not restricted to anatomical boundaries. There is a functional architecture of the brain that represents the work that the brain computes across the the anatomical hardware. This project aims to disentangle the functional architecture by developing new models specialized in exploring the different types of brain computations.

As opposed to current models, the emphasis on finding circuitry within the brain that combines in different ways to support flexible computation will uncover how brains can flexibly adapt to produce robust intelligence.

Specifically, this project will develop this framework to understand how computation is distributed in two important model behaviors: navigation in whole-brain recordings of zebrafish larvae, and decision making in rodents. Aim 1 will identify systems underlying navigation, using the advanced optical imaging tools available that capture the simultaneous activity of both neurons and glia over the entire zebrafish larval brain.

First, a behaviorally-decomposed dynamical system will be developed to explicitly tie the subsystems across the brain to behavioral elements including swim bouts and visual stimuli. Second, the decomposed dynamical systems model will be extended to explicitly capture the asymmetric nature of neural-glial interactions. Aim 2 will identify subsystems underlying visual decision making in multi-electrode recordings of rodents.

Despite the ability to record many neurons in multiple brain areas, these neurons still represent only a fraction of the neurons in these areas, let alone the entire brain. To address this limitation this project will develop a multi-scale missing-neuron dynamical system that will treat hidden systems as a hidden set of latent variables that operate in parallel to the observed neurons.

These methods will reveal how the functional architecture is structured and how different pathways are activated under different conditions, e.g., error trials. Taken together this project will establish new quantitative models to uncover the functional architecture of the brain and provide new insight into both navigation and decision making.

This award reflects NSF's statutory mission and has been deemed worthy of support through evaluation using the Foundation's intellectual merit and broader impacts review criteria.