This project will address a fundamental question both in cognitive psychology and neuroscience: How do the elements of the brain - millions of relatively simple processing units - cooperate to produce purposeful, goal directed behavior, without recourse to components that are imbued a priori with such abilities (i.e., a """"""""homunculus"""""""")? In our previous efforts to address this question, combined computational modeling and empirical studies have lead us to the following hypotheses: a) a cardinal function of prefrontal cortex is to maintain context representations in working memory that guide the flow of activity in more posterior structures responsible for carrying out task-specific processes (e.g., mapping of stimuli to responses); b) the maintenance of these representations relies on attractor system dynamics; c) these representations are selectively updated by the neuromodulatory influence of dopaminergic signals that gate new information into working memory in PFC; and d) the simultaneous role of DA in reinforcement learning trains the network to carry out properly such updating. We have shown that this combination of mechanisms is sufficient to exhibit self-organization, such that the system can learn to execute simple control tasks without recourse to a 'homunculus'. However, important questions remain to be addressed. To date we have focused on updating of only a single representation at a time (i.e., the need to keep track of only a single cue, rule, or goal). However, humans can perform more complex tasks involving hierarchical updating of information (e.g., subgoaling). What additional mechanisms are required to support such selective updating? An equally important question is how the system learns which information must be updated and when. Finally, we know that humans are highly adaptive, humans can use internal monitoring of their own performance, in addition to external feedback, to drive learning. In this project we will extend our existing models to address these questions, and conduct a series of empirical studies to test these models and more general consequences of our theory. Specifically, we will integrate three models that we have developed in previous work, one of which possesses a subset of the required functionality (i.e., selective updating, learning, and performance monitoring), but no one of which exhibits the full set. Success in this work would produce the most comprehensive neurobiologically plausible theory of working memory and cognitive control to date. At the same time, we will conduct a series of fMRI and combined fMRI-ERP studies to test predictions that the model makes about the specific neurobiological mechanisms that underlie these components of cognitive control.
Showing the most recent 10 out of 98 publications
