Optimal accumulation of evidence for decision-making in rats

Abstract

Gradual accumulation of evidence is thought to be a fundamental component of decision-making. However, the mechanisms and properties of the accumulation remain unclear. Although most models have assumed a noisy accumulation process, the properties of this noise have never been isolated and measured. Here we described a decision-making task in which sensory evidence is delivered in brief, isolated pulses whose precisely-controlled timing varies within and across trials. The task is amenable to quantitative analysis that distinguishes between different decision-making dynamics and mechanisms. The task allowed us to measure, for the first time, the magnitude of noise in the accumulator's memory, separately from noise associated with the incoming sensory evidence. Remarkably, we found that the accumulator is essentially perfect: its memory is long (>1 sec) and noiseless. In contrast, the process of adding new sensory evidence is noisy and is the primary source of variability in decision-making. These results put important constraints on mechanisms underlying accumulation of evidence for decision-making.