Skip to main content
. 2016 Jan 8;9:491. doi: 10.3389/fnins.2015.00491

Figure 2.

Figure 2

Representative spike raster plots of classification of a test set at two levels of detail. (A) Detail of spikes occurring during 1 s presentation of a single test digit to the trained Spikey classifier using 10 virtual receptors. The banding discriminates individual populations of 6 or 8 neurons. The colors distinguish the main layers (bottom to top): RN layer (0–59), PN and lateral inhibitory LN neurons (60–129 and 130–189), AN output neurons and paired lateral inhibitory LN neurons (190–205 and 206–222). High activity in the upper AN population determines the classification decision. (B) The trained SpiNNaker classifier using 50 VRs and all 10 digits (0–9). Spiking activity occurring during consecutive 120 ms presentations of 50 × MNIST test digits ordered cyclically 0–9. The colors distinguish the main layers (bottom to top): RN layer (50 clusters of 30 neurons), PN layer (50 clusters of 30 neurons) and at the top, AN output neurons (10 clusters of 30 neurons). A perfectly regular “sawtooth” pattern of activity in the output would imply 100% classification. A similar representative raster plot from GeNN is available as Supplementary Material.