What caused an illness, was it a specific loaf of rye bread or all gluten products? Why did it rain today and is this best explained by the cloud formation, time of year, or both? Such inference problems pervade our daily lives, and solving them adaptively has been critical for survival in our evolutionary past. All humans–from children to scientists–have an exceptional ability to detect and leverage structure present in the world. This capacity dramatically reduces the computational demand imposed by the stream of sensory information we receive daily by enabling us to abstract meaningful information and facilitating generalizations to entirely new situations that share similar structural properties.
Our labs use a combination of behavioral, simulation, and functional imaging techniques to look at the neural networks involved in representing casual systems, as well as the computations that underlie updating and or changing our belief in the validity of the casual relationship. For example, in Boorman et al. (2016) we use an innovative new fMRI technique to show a dynamic medial temporal network which changes as function stimulus-outcome associations, suggesting that it flexibly encodes the relationship between a predictor stimulus and its outcome. Moreover, we found two distinct neural signals that underlie this form of learning. One signal in the Orbitofrontal Cortex was shown to be involved in updating our beliefs in stimulus-outcome associations - the extent to which we believe a predictor leads to a certain type of outcome. This signal was also found to be computationally distinct from another signal in the ventral midbrain which was sensitive to reward value. These results advance our understanding of how learning related signals are related to forming and flabbily encoding neural representations of associative structures.