Identifying positive roles for endogenous stochastic noise during computation in neural systems

Information processing in nonlinear systems can sometimes be enhanced by the presence of stochastic fluctuations, or noise. Although the electrical properties of neurons and synapses are known to be influenced by intrinsic stochastic variability, it remains an open question as to whether living systems exploit this noise during neuronal information processing. This is despite various forms of noise-enhanced processing, such as classical stochastic resonance, having been observed in mathematical models of neural systems and in data acquired experimentally. We recently argued that advancing our understanding of the potential roles of random noise in assisting neuronal information processing will require specific focus on a concrete hypothesis about the computational roles of a specific neural system that can then be tested experimentally using signals and metrics relevant to the hypothesis. In this invited symposium paper, we argue why most existing approaches to studying stochastic resonance based on classical definitions and methods are highly limited in their applicability, since they impose an implied computational hypothesis that may have little relevance for real neurobiological systems.