Figure 11.

Average angle accuracy for an increasingly sparse synaptic delay network. We connected each LIF neuron to a random subset of 8 pre-synaptic neurons representing the hardware sensors. We varied the size of the subset across 35 different trials in order to assess the smallest number of connections between pre-synaptic and LIF neurons capable of preserving an accurate temporal representation of all 8 angles. A sparse synaptic delay network with only 2 connections per LIF can already represent all angles with an average accuracy of 95% and we can achieve 100% accuracy with only 4 connections.