Fig. 6
From: Attractive and repulsive effects of sensory history concurrently shape visual perception
Generative model. A Tuning curves of direction selective neurons after adaptation at 180° (black arrow). Adaptation was modeled as a reduction of the response gain in neurons selective to the adapter direction. B Population likelihood. Likelihoods were computed assuming independent Poisson spike count variability and averaged over 1,000 presentations of the stimulus direction θt, each following a presentation of the adapter direction as in A. Due to adaptation, the likelihoods are on average shifted away from the adapter direction. Shaded regions represent ±1 SD. C Recurrent Bayesian inference. The model first convolves posterior belief on the previous stimulus p(θt − 1| r1 : t − 1) with the propagation noise distribution p(θt| θt − 1), to predict the current stimulus θt based on the past sensory responses r1 : t − 1. The model then combines this prediction p(θt| r1 : t − 1) with the new sensory information p(rt| θt), to produce posterior belief on the current stimulus p(θt| r1 : t). D Model simulation. Simulation data were analyzed and plotted as in Fig. 2A. The model successfully generated the characteristic features of the human estimation behavior