Event-based computation:Unsupervised elementary motion decomposition

Fast, localised motion detection is crucial for anefficient attention mechanism. We show that modelling a networkcapable of such motion detection can be performed using spikingneural networks simulated on many-core neuromorphic hardware. Moreover, highly sensitive neurons arise from the presentednetwork architecture through unsupervised self-organisation. Weuse a synaptic rewiring rule which has been shown to enablethe formation and refinement of neural topographic maps. Ourextension allows newly formed synapses to be initialised with adelay drawn from a uniform distribution. Repeated exposure tomoving bars enables neurons to be sensitised to a preferred direction of movement. Incorporating heterogeneous delays results inmore sensitive neural responses. A readout mechanism involvinga neuron for each learnt motion is sufficient to establish the inputstimulus class