In our newest study on spike-based Bayesian computation, we address the question of well-calibrated stochasticity in deterministic spiking systems. We show that even a small reservoir of excitatory and inhibitory neurons can provide a much larger ensemble of functional spiking networks with the required pseudo-randomness without skewing its sampling statistics. The main underlying mechanism lies in the cancelling out of positive shared-input correlations in the functional networks by the negative cross-correlation of spiking activity in the reservoir. This represents a particularly efficient mechanism for pseudo-randomness in both biological and artificial neural networks.
The publication is open access and can be found here: https://doi.org/10.1038/s41598-019-54137-7
Last update of this page: 2019-12-09 by A. Baumbach
HOME
