Biologically plausible learning with spiking neural networks : Image classification with STDP-based learning rules : Bildkassifizierung mit STDP-basierten Lernregeln

Abstract

In the last decade, deep learning architectures such as artificial neural networks (ANN), convolutional neural networks (CNN) and recurrent neural networks (RNN) have been dominating in several fields. These state of the art deep neural networks, however, represent only a very simplified version of their biological role model. More recently, alternative architectures such as Spiking Neural Networks (SNN) have been explored that try to more precisely model neural processing in the brain. These are inspired by the communication of biological neurons, especially the temporal information transformation via discrete action potentials, or spikes, through adaptive synapses and allow asynchronous information processing. One promising approach that allows unsupervised learning of the synaptic weights in SNNs is based on the principle of Spike-Timing Dependent Plasticity, or STDP. The investigation of this alternative network architecture together with STDP-based learing rules represents an opportunity to derive new insights about the robustness and energy-efficient implementations of neural network in specialized hardware and also in general.