Extended Data Fig. 9: Description and identifiability of the two Bayesian observer models.
a, Schematic illustration of the Bayesian observer model that combines the prior (top left) with the likelihood function (top right: three example likelihood functions associated with three intervals) and uses the posterior mean to produce the desired interval. We considered two noise models for interval production. In one model (left), the standard deviation of noise scales with temporal distance. This model is consistent with path integration without incorporating landmarks resets. In the other model (right), the standard deviation increases sublinearly with temporal distance (variance increases linearly with temporal distance). This model is consistent with mental navigation with landmark resets. b, Distribution of mean squared error (MSE) between ground truth data generated from a chosen generative model and data generated from the two models fitted to the ground truth data. Left: ground truth data generated from a model with reset (blue). Right: ground truth data generated from a model without reset (red). c, Distribution of fitted parameter values (Weber fraction, wp) of the two models fitted to data generated by the model with reset (left) and to data generated by the model without reset (right).
