Explore a lecture on cortical area interactions through a communication subspace, delivered by Christian Machens from Champalimaud Research at the Simons Institute. Delve into the targeted discovery of brain data, examining linear readouts from populations, task parameter dependencies, and the decomposition of PSTHs into marginalized averages. Investigate condition-independent components and inter-area communication theory, focusing on simultaneous recordings in V1 output layers and V2 input layers of macaques. Analyze the relationship between V1 population activity and single V2 neuron responses, as well as V2 population predictions. Examine the dimensionality of V1-V2 interactions, predictive dimensions, and their relationship to dominant dimensions in neural activity. Gain insights into how prediction of target V1 activity relies on source V1 dominant modes in this comprehensive exploration of cortical communication.
Cortical Areas Interact through a Communication Subspace
Overview
Syllabus
Intro
Linear readout from a population
(1) demix dependencies on task parameters
Algorithm Step 1: decompose PSTHS into marginalized averages
Condition-independent components: Only part of manifold is experimentally controlled
Inter-area communication: theory
Inter-area communication: experiment
Simultaneous recordings in V1 output layers and V2 input layers of macaque
Population activity quantified as spike counts
Relating V1 population activity to response of a single V2 neuron
Predicting the V2 population Could predictive dimensions span a subspace?
Control: Is V2 activity lower dimensional than V1 activity?
V1-V2 interaction is low dimensional
Controls
Do predictive dimensions match dominant dimensions?
Predictive dimensions do not match dominant dimensions
prediction of target V1 activity relies on source V1 dominant modes
Taught by
Simons Institute
