The computational task of the brain is to process noisy signals of an ambiguous environment with a stochastic hardware. The mathematical framework to address this task is bayesian statistics which allows to represent and compute with uncertain quantities. Finally, this computation must be implemented with neurons and synapses. Linking these three levels, i.e. computational task, mathematical representation and implementation can be achieved by normative models.
In my PhD, I propose a normative model to explain the phenomenon of synaptic stochasticity from a computational perspective. Synaptic stochasticity can be mechanistically understood in terms of stochastic vesicles release. Why study it? Because it might actually have a functional role in representing model-uncertainty during brain computation.
My normative model is based on the assumption that the brain makes predictions based on averages over model parameters, each represented by a weight distribution, instead of just taking the best one, i.e. the brain carries out bayesian regression. To this end, I am working out experimental predictions of Bayesian regression on the psychophysical level as well as the level of synaptic plasticity. In particular, I address the question of how the brain would implement the learning and prediction task based on weight distributions.
Constrained basin stability for studying transient phenomena in dynamical systems
Authors: A. van Kan, J. Jegminat, J. Donges, J. Kurths.
Phys. Rev. E
doi = 10.1103/PhysRevE.93.042205
url = http://link.aps.org/doi/10.1103/PhysRevE.93.042205