Effects Of Real-time Synaptic Plasticity Using Spiking Neural Network Architecture

Artificial Neural Networks is a promising approach to study human brain computation in hopes of achieving similar learning by artificial agents. Recent architecture design of a low-power supercomputer by the University of Manchester, the SpiNNaker, has made it easier to design highly parallel brain-inspired algorithms. We used the SpiNNaker machine to implement a neural network capable of rewiring its connection in real-time while trying to minimize information loss and maximize the decrease in statistical dependence. We believe this is the first use of synaptogenesis on a spiking neural network architecture, laying the framework for future efficient brain-like neural networks. Additionally, we explored scalability issues and unintended pitfalls with this approach. Keywords—Neural Networks, Synaptogenesis, SpiNNaker