While recent years have seen rapid advances in our understanding of the network of brain regions associated with cognitive control and decision making, the precise function of these regions remains highly ambiguous. A single region may appear to perform several different functions depending on the experimental task as well as the methodology used to measure neural activity. One particularly
important region in this respect is medial prefrontal cortex (mPFC). In recent months, two groups working independently have published computational models of mPFC that account for a range of effects observed within the region under different experimental tasks and measurement methodologies. The formulation of the predicted response-outcome (PRO; Alexander & Brown,
2011) and the reward value and prediction (RVPM; Silvetti, Seurinck & Verguts, 2011) models is informed by reinforcement learning theory, and interprets mPFC function as signaling predictions of future outcomes as well as deviations from those predictions. These models have an unprecedented ability to comprehensively account for activity in mPFC based on well-known reinforcement learning formalizations. This suggests that the network of regions involved in cognitive control, including mPFC but also other areas such as dorsolateral prefrontal cortex, may implement some form of reinforcement learning. This project will test this hypothesis through a combined program of empirical investigation using fMRI and computational modeling based on our recently published
work.