It is becoming increasingly evident that organisms acting in
uncertain dynamical environments employ exact or approximate
Bayesian statistical calculations in order to continuously
estimate the environmental state and integrate information from
multiple sensory modalities. What is less clear is how these
putative computations are implemented by cortical neural networks.
We show how optimal real-time state estimation based on
noisy multi-modal sensory information may be effectively implemented by neural networks decoding sensory spikes. We demonstrate the efficacy of the approach on static decision problems as well as on dynamic tracking problems, and relate the properties of optimal tuning curves to the properties of the environment.