Uncertainty propagation in neuronal dynamical systems

One of the most notorious characteristics of neuronal electrical activity is its variability, whose origin is not just instrumentation noise, but mainly the intrinsically stochastic nature of neural computations. Neuronal models based on deterministic differential equations cannot account for such variability, but they can be extended to do so by incorporating random components. However, the computational cost of this strategy and the storage requirements grow exponentially with the number of stochastic parameters, quickly exceeding the capacities of current supercomputers. This issue is critical in Neurodynamics, where mechanistic interpretation of large, complex, nonlinear systems is essential. In this paper we present accurate and computationally efficient methods to introduce and analyse variability in neurodynamic models depending on multiple uncertain parameters. Their use is illustrated with relevant examples