Bump attractor dynamics underlying stimulus integration in perceptual estimation tasks

Abstract

Perceptual decision and continuous stimulus estimation tasks involve making judgments based on accumulated sensory evidence. Network models of evidence integration usually rely on competition between neural populations each encoding a discrete categorical choice and do not maintain information that is necessary for a continuous perceptual judgement. Here, we show that a continuous attractor network can integrate a circular stimulus feature and track the stimulus average in the phase of its activity bump. We show analytically that the network can compute the running average of the stimulus almost optimally, and that the nonlinear internal dynamics affect the temporal weighting of sensory evidence. Whether the network shows early (primacy), uniform or late (recency) weighting depends on the relative strength of the stimuli compared to the bumps amplitude and initial state. The global excitatory drive, a single model parameter, modulates the specific relation between internal dynamics and sensory inputs. We show that this can account for the heterogeneity of temporal weighting profiles and reaction times observed in humans integrating a stream of oriented stimulus frames. Our findings point to continuous attractor dynamics as a plausible mechanism underlying stimulus integration in perceptual estimation tasks.