A solution to the learning dilemma for recurrentnetworks of spiking neurons

Guillaume Bellec, Franz Scherr, Anand Subramoney, Elias Hajek, Darjan Salaj, Robert Legenstein, Wolfgang Maass

Abstract: Recurrently connected networks of spiking neurons underlie the astounding informationprocessing capabilities of the brain. Yet in spite of extensive research, how they can learnthrough synaptic plasticity to carry out complex network computations remains unclear. Weargue that two pieces of this puzzle were provided by experimental data from neuroscience.A mathematical result tells us how these pieces need to be combined to enable biologicallyplausible online network learning through gradient descent, in particular deep reinforcementlearning. This learning method–called e-prop–approaches the performance of back-propagation through time (BPTT), the best-known method for training recurrent neuralnetworks in machine learning. In addition, it suggests a method for powerful on-chip learningin energy-efficient spike-based hardware for artificial intelligence.