Integrating Spiking Neural Networks and Deep Learning Algorithms on the Neurorobotics Platform

Rachael Stentiford, Thomas C. Knowles, Benedikt Feldoto, Deniz Ergene, Fabrice O. Morin, Martin J. Pearson

Abstract: We present a neurorobotic model that can associate self motion (odometry) with vision to correct for drift in a spiking neural network model of head direction based closely on known rodent neurophysiology. We use a deep predictive coding network to learn the generative model of representations of head direction from the spiking neural network to views of naturalistic scenery from a simulated mobile robot. This model has been deployed onto the Neurorobotics Platform of the Human Brain Project which allows full closed loop experiments with spiking neural network models simulated using NEST, a biomimetic robot platform called WhiskEye in Gazebo robot simulator, and a Deep Predictive Coding network implemented in Tensorflow.