Dopaminergic and frontal signals for decisions guided by sensory evidence and reward value [article]

Armin Lak, Michael Okun, Morgane Moss, Harsha Gurnani, Miles J Wells, Charu Bai Reddy, Kenneth D Harris, Matteo Carandini
2018 bioRxiv   pre-print
Making a decision often requires combining uncertain sensory evidence with learned reward values. It is not known how the brain performs this combination, and learns from the outcome of the resulting decisions. We trained mice in a decision task that requires combining visual evidence with recent reward values. Mice combined these factors efficiently: their decisions were guided by past rewards when visual stimuli provided uncertain evidence, but not when they were highly visible. The sequence
more » ... f decisions was well described by a model that learns the values of stimulus-action pairs and combines them with sensory evidence. The model estimates how sensory evidence and reward value determine two key internal variables: the expected value of each decision and the prediction errors. We found that the first variable is explicitly represented in the activity of neuronal populations in prelimbic frontal cortex (PL), which occurred during choice execution. The second variable was explicitly represented in the activity of dopamine neurons of ventral tegmental area (VTA), which occurred after stimulus presentation and after choice outcome. As predicted by the model, optogenetic manipulations of dopamine neurons altered future choices mainly when the sensory evidence was weak, establishing the causal role of these neurons in guiding choices informed by combinations of rewards and sensory evidence. These results provide a unified, quantitative framework for how the brain makes efficient choices when challenged with internal and environmental uncertainty.
doi:10.1101/411413 fatcat:ymoeedwx6bbxjmdm7xmx6mdxk4