The Synaptic Kernel Adaptation Network

In this paper we present the Synaptic Kernel Adaptation Network (SKAN) circuit, a dynamic circuit that implements Spike Timing Dependent Plasticity (STDP), not by adjusting synaptic weights but via dynamic synaptic kernels. SKAN performs unsupervised learning of the commonest spatio-temporal pattern of input spikes using simple analog or digital circuits. It features tunable robustness to temporal jitter and will unlearn a pattern that has not been present for a period of time using tunable 'forgetting' parameters. It is compact and scalable for use as a building block in a larger network to form a multilayer hierarchical unsupervised memory system which develops models based on the temporal statistics of its environment. Here we show results from simulations as well present digital and analog implementations. Our results show that the SKAN is fast, accurate and robust to noise and jitter.