skip to main content

Behavioral Social Neuroscience Seminar

Friday, October 31, 2014
4:00pm to 5:00pm
Add to Cal
Beckman Behavioral Biology B180
Neural Computations Central to Learning What to do in an Uncertain World
Erie Boorman, Sir Henry Wellcome Postdoctoral Research Fellow, University of Oxford,

Making good choices requires people to appropriately revise their beliefs in the face of new information. This information can originate from many sources - e.g. direct experience, counterfactuals, observation - and take many forms - e.g. reward value, reward identity, another person's attributes. In a series of experiments, I have investigated the behavioral and neural correlates of prediction errors, the engines of learning, in this variety of settings and forms. In the context of value learning, I will present evidence that simultaneous prediction error signals occur in ventral striatum/ventral midbrain and frontal polar cortex/dorsomedial frontal cortex that respectively update how valuable selected options are and counterfactual options would have been had they instead been selected. I will also present recent findings showing that both single neurons and the population of neurons in rostral ACC encode observational prediction errors when people learn vicariously. While learning the value of experienced, foregone, and observed options is of clear evolutionary importance, another crucial question in reinforcement learning concerns how people learn the likely transitions from their choices to specific outcome identities. Such 'goal-directed learning' should enable people to predict what kind of outcome will likely result from their choices in order to flexibly construct value expectations that guide choices. Using a new approach termed cross stimulus suppression, I will first show that hippocampus and amygdala rapidly encode associations between specific stimuli and specific reward outcomes. Using computational fMRI, I will then show that simultaneous reward value and reward identity updates can be measured in ventral midbrain and lateral OFC. Combining the two approaches suggests that independent learning pathways from ventral midbrain and lateral OFC converge in the hippocampus where they drive rapid changes to stimulus-reward associations.  Finally, a similar contingent learning mechanism involving lateral OFC appears to be recruited to update associations between specific outcomes and specific other agent's choices during social attribute learning.  Collectively, these findings endorse the view that sequential behavior in humans is guided by multiple learning systems and validate a new technique that shows promise for probing the dynamics of other learning and representation questions.

For more information, please contact Jenny Niese by phone at Ext. 6010 or by email at [email protected].