Project Summary/Abstract
There is considerable evidence that the brain relies on a set of canonical neural computations, repeating them
across brain regions and modalities to apply operations of the same form, but we lack a theoretical framework
for how such canonical neural circuit computations can support a wide variety of cognitive processes and brain
functions. Through the proposed research, we aim to provide one. Preliminary results demonstrate that a
family of circuit models, called Oscillatory Recurrent Gated Neural Integrator Circuits (ORGaNICs), simulates
many key neurophysiological and cognitive/perceptual phenomena. We propose to develop models of the
dynamics of attention, and working memory, and to test those models with previously published datasets
acquired with a wide range of methodologies: human behavioral data, neurophysiological data from primate
and rodent prefrontal cortex (PFC), electrophysiology and Ca2+ imaging data from rodent prefrontal cortex,
and electrophysiology data from rodent medial entorhinal cortex.
In Aim 1, we hypothesize that normalization is critical for the stability and robustness of the recurrent circuits
that underlie working memory. We will test this hypothesis by developing an analytical theory, based on
ORGaNICs, of delay-period activity, and fitting published measurements of response dynamics in PFC.
In Aim 2, we hypothesize that behavioral performance during working memory tasks is limited by trial-to-trial
variability in delay-period activity, and also that top-down signals from working memory circuits provide the
attention-control signals that modulate sensory activity in visual cortex. We will test these hypotheses by
developing an analytical theory of attention and working memory, combining a visual cortex model and a PFC
model, and using it to fit previously published measurements of behavioral performance from a variety of
attention and working-memory experiments.
In Aim 3, we propose to develop and test a theory of manipulation in working memory, with application to
navigation, specifically using ORGaNICs to model the responses of populations of head-direction cells while
animals are performing the active place avoidance task. We hypothesize that head-direction cells in MEC
operate like a working memory representation, by encoding a “landmark” (a sensory feature) relative to the
animal's current head direction, and then updating/manipulating the representation of that landmark as the
animal's orientation changes. We will test key predictions of the theory with new experiments.
The proposed research has the potential to be transformative by: providing a new set of analytical results and
computational (software) tools for characterizing and simulating a broad range of neural circuit models, which
will impact experimental design and data analysis; making new experimentally-testable predictions for both
ORGaNICs and alternative models; and testing some of those predictions with new experiments.