Speakers > Alfonso Renart

Alfonso Renart

Champalimaud Institute, Lisbonne, Portugal

I studied physics and did my PhD in Computational Neuroscience, both at the Universidad Autónoma de Madrid. After a postdoc at Brandeis University — in the lab of Xiao-Jing Wang — and another one at Rutgers University — in the lab of Kenneth Harris — I became group leader of the Circuit Dynamics and Computation Lab at the Champalimaud Neuroscience Programme. In the lab, we combine theoretical and experimental approaches, and careful behavioral analysis, to study the computational principles underlying perceptual decision-making.

The mechanics and norms of perceptual choice

Work over the last 30 years in perceptual decision-making has identified bounded accumulation of evidence as a core computational principle describing how animals use information from the environment to guide categorical choices. In this seminar, I will describe two recent developments in our lab that characterize decision-making from a mechanistic and from a normative perspective. First, I will consider the problem of perceptual choice, but taking into account not only the relative evidence in favor of each alternative, but also the overall sensory intensity across alternatives — thus connecting modern approaches in decision-making with foundational questions in psychophysics. Mathematical analysis of behavioral experiments in rats and humans allowed us to identify the mechanistic basis of Weber’s law, and to show that it follows naturally from a bounded accumulation of evidence framework. In the second part of the talk, I will describe a normative view on perceptual choice, that is, how agents should use sensory information to guide decisions. I will argue that the natural cost function to be optimized should include not only performance costs — as is generally assumed — but also costs derived from controlling default behavioral strategies shaped by evolution or prior learning, but possibly maladaptive in the context of a laboratory experiment. Using this philosophy, we derived optimal policies for perceptual choices in control-limited agents. These policies rely on accumulation of evidence, but control-limitations lead to probabilistic decision bounds. Our work shows that considering different levels of cognitive control explains a range of phenomena in decision-making, and provide a path for studying optimal decision strategies in real biological agents.