Figure 1.

Network architecture, connectivity matrix, and general layout of a recurrent network. A, Circuit diagram of the fly motion vision system. Retinal input (gray, only one is shown) arrives at laminar neurons (green), which project to medulla (red) and lobula (blue) neurons. From there, activity reaches the T4 (dark red) and T5 (dark blue) cells. In this system, the action of the neurotransmitters is known: arrows indicate excitation; discs indicate inhibition. B, The number of synapses between neuron types is summarized in a connectivity matrix. C, If one interprets the number of synapses between each pair of cells as the total synaptic strength of this connection, the matrix in B defines a recurrent neural network in which inputs from all presynaptic cells are collected at “dendrites” (horizontal bars). The retinal input I(t) provides additional external drive to each neuron. A, B, Reproduced from Borst et al. (2020b), with permission.