Bayesian head state prediction: Computing the dynamic prior with spiking neurons

In a model of the vestibular system the Bayesian posterior density of head state is represented by the spatial density of spikes in brainstem vestibular neurons. Individual sensory spikes are measurements of head state. We show how to compute the Bayesian dynamic prior density for each sensory spike, from the posterior density based on previous spikes, using natural neuron-like operations on spikes. Head movement dynamics together with Bayes rule determine the architecture of the required neural network. This is a natural model of neural computation using spikes as operands.