Our work in collaboration with researchers from the University of Bern, "Stochasticity from function - why the Bayesian brain may need no noise", has been published in the journal "Neural Networks". Inspired by the modular structure of the cortex, we show that spike-based sampling can be implemented in ensembles of functional networks that provide their own background noise, without explicit noise sources like Poisson input. In this framework, spikes take on a dual role: as samples from a posterior probability distribution, performing Bayesian inference, and as a source of irregularity, enabling sampling in the first place. This way, robust spike-based stochastic computing can be realized in a completely self-contained and deterministic system, which we demonstrate in software simulations as well as on the BrainScaleS "physical model" system.
The publication is open access and can be found here: https://doi.org/10.1016/j.neunet.2019.08.002.
Last update of this page: 2019-08-26 by Dominik Dold
HOME
