Fig. 2: Simulated spiking neural network with a biologically inspired architecture.

a Network schematic. In the mammalian sensory cortex, excitatory and inhibitory neurons are laterally connected with higher connection probabilities for nearby neurons. We simulated a 2D sheet populated by interspersed excitatory and inhibitory spiking neurons in the same ratio found in the cortex (4:1). A specific sensory stimulus (e.g., an auditory tone) provides excitatory thalamic inputs to a fraction of cells (neurons selective to the tone’s frequency; shaded region at the center of the sheet). b Probability of synaptic connection between different types of cells as a function of spatial distance of neurons. The x-axis represents the distance between nodes measured in normalized units (i.e., distance of two horizontal or vertical adjacent cells is 1), and distance of two diagonal adjacent cells is \(\sqrt{2}\). c Weight of synaptic connections between different cell types as a function of spatial distance. d Temporal profile of thalamic excitation following the stimulus onset (time 0) received by cells in different locations of the 2D sheet. e Raster plots of representative excitatory and inhibitory neurons shortly after the stimulus onset and at a later time when thalamic inputs have subsided to baseline. Neurons continue to generate spikes after the stimulus due to network reverberations. f Heatmap of spike counts of excitatory cells on the 2D sheet. The central square with more pronounced spiking corresponds to the sub-region directly excited by the thalamic inputs (shaded region in a). g Spiking activity of simulated neurons is stochastic. The average Fano factor of spike counts of excitatory cells is slightly above one before the stimulus onset, and briefly plummets after the stimulus onset, matching past experimental observations (e.g.,¹⁵). Fano factors calculated with sliding windows of different lengths show similar trends.