Figure 1.

Deep Continuous Local Learning (DECOLLE). (Left) Each layer consists of spiking neurons with continuous dynamics. Each layer feeds into a local readout units through fixed, random connections (diamond-shaped, y). The local layer is trained such that the readout at each layer produce auxiliary targets Ŷ. Errors are propagated through the random connections to train weights coming into the spiking layer, but no further (curvy, dashed line). To simplify the learning rule and enable linear scaling of the computations, the cost function is a function of the states in the same time step. The state of the spiking neurons (membrane potential, synaptic states, refractory state) is carried forward in time. Consequently, even in the absence of recurrent connections, the neurons are stateful in the sense of recurrent neural networks. (Right) Snapshot of the neural states illustrating the DECOLLE learning rule in the top layer. In this example, the network is trained to produce three time-varying pseudo-targets Ŷ¹, Ŷ², Ŷ³.