Skip to main content
Figure 1 | BMC Neuroscience

Figure 1

From: Inhibitory interneurons enable sparse code formation in a spiking circuit model of V1

Figure 1

A. Circuit diagram of our spiking network with separate excitatory (E) and inhibitory (I) neural populations (top) compared to current single population models (bottom). This network was simulated with different numbers of excitatory and inhibitory cells. B. Adding inhibitory cells to the network differentiates the receptive fields and decreases image reconstruction error during learning. C. This error reduction is caused by decreased correlations among the excitatory neurons that are collaborating to form a sparse representation of the visual input. The network was trained on 8x8 image patches (64 pixels) drawn from whitened natural images. Excitatory neuron counts (# E cells) ranged from 64 to 384 (1x to 6x overcomplete). Inhibitory neuron counts (# I cells) ranged from 3 to 64 (.05x to 1x overcomplete). We find that reconstruction errors are roughly constant for populations of interneurons that are at least ~20% of the size of the total population, assuming the total neural population is at least 4x overcomplete relative to the input. This is consistent with the 80/20 ratio of excitatory-to-inhibitory neurons observed in visual cortex.

Back to article page