Cognitive dynamics during formation and revision of perceptual decisions

Abstract

A central question in neuroscience is how the nervous system integrates information available in the environment to guide ongoing behavior. What are the algorithms governing this process and how are they implemented in neural circuits? In this thesis, I present three studies that address these questions within the paradigm of perceptual decision-making.In the first, we reanalyzed a previous study of neural responses in two rat cortical regions during evidence accumulation. We found that while the average dynamics of choice-selective neurons contain evidence-dependent ramping toward a fixed activity level at the time of choice, individual cells have reliable peaks of activity throughout the trial that span every stage of the decision. In the second study, we trained rats to perform a dynamic decision task designed to induce “changes of mind,”—fluctuations in the subjects' provisional decisions—throughout deliberation. We used a latent variable model of the decision process, fit to each rat's choices, to estimate the timing of these change of mind events, and asked whether these events could account for changes in neural activity in a secondary motor cortical region called the Frontal Orienting Fields (FOF). We found that these model-predicted changes of mind were well-aligned to moments when neural activity changed across the population of choice-selective cells, indicating that FOF neurons encoded the rat's provisional decision even as it changed to reflect the new state of the environment. In the third study, we combined an evidence integration decision task with a paradigm for measuring decision confidence in which rewards are randomly delayed and subjects must decide how long to wait before giving up and starting a new trial. We developed a new model for describing the evolution of a decision variable that produces the wait time decision. The model unifies the initial choice with the wait time decision and captures the distribution of wait times we observed in rats. This model will improve the interpretive power of this paradigm by providing an evolving decision variable that can be studied alongside neural recordings.