Bio-Inspired Computation: Clock-Free, Grid-Free, Scale-Free and Symbol Free.

摘要

The project developed a new fundamental component for bio-inspired computing, based on a new way of modelling spiking neurons, and applying them to a new type of long-range temporal prediction task. The new model neuron has been applied to event-based data from a new type of motion sensitive camera - the neuromorphic Dynamic Vision Sensor (DVS). The model neuron incorporates temporal delays on both dendrites (inputs to neurons) and axons (outputs from neurons). Delays on axons have not previously been modeled in sensory-motor processing tasks, and their addition significantly simplifies asynchronous network development with spiking neurons, in particular reducing the computational complexity of algorithms for sparse data over dense sensory arrays. Effectively, the new model neuron treats each spike as a connection between temporal patterns extended in time in both its past and future.