For decades now, normative theories of perceptual decisions, and their implementation as drift diffusion models (DDMs), have driven and significantly improved our understanding of human and animal behavior and the underlying neural processes. While similar processes seem to govern value-based decisions, we have lacked the theoretical understanding of why this ought to be the case. Recently, we have mathematically proved that, similar to perceptual decisions, DDMs implement the optimal strategy for value-based decisions. Such optimal decisions require the models’ decision boundaries to collapse over time, and to depend on the a priori knowledge about reward contingencies . Here, we further extend the theory to more general cases involving multiple choice alternatives, revealing the optimal strategy for such complex decisions as well as a possible neural circuit to implement the optimal strategy. Crucially, the neural implementation involves a nonlinearity known as normalization which had been previously blamed for so-called irrational behaviors in value-based decision making. Our work instead demonstrates that value normalization is not a bug but a feature of the optimal policy. Together, our findings provide the much-needed theory for value-based decisions, explain the apparent similarity to perceptual decisions, and suggest that the apparently “irrational” choice behavior resulting from the activity normalization is a necessary component of the optimal decisions.
 Tajima, S., Drugowitsch, J., & Pouget, A. (2016) Optimal policy for value-based decision-making. Nature Communications, 7:12400.
Satohiro (“Sato”) Tajima received the Ph.D. degree in Engineering from the University of Tokyo in 2013, during working at Japan Broadcasting Corporation. He has been studying theoretical neuroscience and cognitive science as a postdoc at the Department of Basic Neuroscience, University of Geneva, Switzerland since 2014. He is the recipient of William James Prize in 2015 from Association for the Scientific Study of Consciousness. He is currently leading a project supported by PRESTO, JST, "Collaborative Mathematics for Real World Issues," (2016-2020) as an independent investigator. His research area includes vision, natural statistics, decision making, and nonlinear dynamics.
Neuroscience Special Seminar by the NYU-ECNU Institute of Brain and Cognitive Science at NYU Shanghai