Deep learning with dense random neural networks

Abstract

We exploit the dense structure of nuclei to postulate that in such clusters, the neuronal cells will communicate via soma-to-soma interactions, aswell as through synapses. Using the mathematical structure of the spiking Random Neural Network, we construct a multi-layer architecture for Deep Learning. An efficient training procedure is proposed for this architecture. It is then specialized to multi-channel datasets, and applied to images and sensor-based data.