The oscillatory dynamic link matcher for spiking-neuron-based pattern recognition

In this paper we show that an unsupervised two-layered oscillatory neural network with inter- and intra-layer connections, and a learning rule based on stimulus difference can behave as a dynamic link matching machine for invariant pattern recognition. We show that this architecture is robust to affine transformations. We call this architecture oscillatory dynamic link matching (ODLM). We use DEVS (discrete-event simulation) to increase simulation speed by updating the network only at instants when an internal or external stimulus is applied to neurons.1