Project Overview

Rodent models offer an opportunity to understand the computations performed in neural circuits at a much precise scale than using human neuroimaging. We can record from hundreds of identified individual neurons while manipulating neural activity with millisecond precision. These technologies allow us to obtain a much deeper understanding of the mechanisms underlying the interaction of different neuron types in mammalian brains.

One challenge, however, is to have behaviors that are complex enough to require complex cognitive computations, such as estimating confidence to guide future behavior, but simple enough that rodents can be trained to perform them. This proposal will push the frontier of task complexity to design a task where rats will make decisions based on uncertain sensory information. Rats will then invest time into their decision, obtaining a larger reward for time invested for correct decisions, but a larger punishment if the decision was incorrect. This paradigm, will allow us to measure both a binary decision and a graded report of confidence on every single trial, allowing to probe with high precision, the underlying neural computations. We will also develop innovative surgery techniques, to record simultaneously from 4 brain areas while rats are performing this task. This will allow us to understand how value and confidence representations across areas are modulated by the learning signal from dopamine neurons. Finally, we will develop computational tool to build a data analysis pipeline that leverages recent advances in machine learning to analyze the dynamical structure of neural representations across areas. Together, these contributions will provide foundational new behavioral paradigms and computational tools to understand the neural computations underlying the sense of confidence. They will be used broadly within the neuroscience community to understand how the heterogeneity of neurons in the brain contributes to cognition in health and disease.

Principal Investigator

Paul Masset , The Royal Institution for the Advancement of Learning/McGill University