Speaker:
Bruno Averbeck, National Institute of Mental Health, Bethesda, USA
Abstract:
Drift diffusion processes have been used to model choice behavior across diverse tasks. Whether these models can describe neural mechanisms in tasks beyond perceptual decision making (PDM), where they have been used extensively, is an open question. In the present study we examined neural activity in lateral prefrontal cortex (area 46) and the dorsal striatum (caudate nucleus) while monkeys carried out interleaved blocks of PDM and reinforcement learning (RL). To examine the decision processes at a neural level, we decoded the animals choices from single neuron activity, and used the decoding analysis to estimate the animals belief about which choice was correct in a time resolved way. In drift diffusion models information is integrated across time. Thus, when there is more information available at each point in time, it accumulates faster. We found in both tasks that the mean belief about which choice was correct increased with time and reflected choice difficulty, consistent with a drift diffusion process. Variability in information accumulation should also depend on choice difficulty. We found, consistent with this, that variability in belief estimates was higher in conditions where choices were more difficult. Thus, as the animals accumulated information about correct choices, the mean and variance of neural activity reflected information accumulation, consistent with a drift diffusion process. Further, this was true in both the PDM task and the reinforcement learning task. We also estimated the time-constant of the neural integrator and found that it was longer in the PDM task than it was in the RL task. Finally, belief at the time of decision correlated with accuracy in the PDM task but not in the RL task. Thus, neural responses were consistent with an integrator in both tasks, but in the RL task the time constant of integration was shorter and belief and choice accuracy were uncorrelated.