Figure 5.

Simulations of the SVV drift and aftereffect during and after prolonged head tilt (dashed red lines) by the Bayesian spatial-perception model. The top panels show changes in the individual parameters of the model over time and the bottom panels show the SVV drift produced by the model in each scenario (blue lines). For these sample simulations, we used the parameter values from subject SR in Table 2, De Vrijer et al. (2009). (A) Head-in-space sensory input ($\widehat{{\text{H}}_{\text{S}}}$): sensory inputs typically decrease over time in the presence of a constant stimulus and there is often an opposite response after the stimulus is removed. In this context, the decay in head-in-space sensory inputs during head tilt would result in the drift of SVV in the direction of the head tilt. After the head returns to upright position, the opposite sensory response would cause the aftereffect. This scenario is consistent with the actual SVV drift and aftereffect in our data. (B) The head position prior (H_Sp): the prior might gradually drift towards the actual head position during head tilt. In this case, however, the SVV drift from the model would be in the opposite direction of the head tilt and it does not match the actual pattern of SVV drift. (C) Increase of the sensory noise (α₁): the head-in-space sensory estimate may become less reliable over time (i.e., noisier), which would gradually increase the weight of the head prior in SVV responses. This scenario also produces an SVV drift toward the actual head position. However, it could not account for the SVV aftereffect, because when the head returns to upright, the noise will reduce to the baseline (α₀), irrespective of the change in the value of α₁ (see Equation 2).