US 12,147,902 B2
Backpropagation of errors in pulsed form in a pulsed neural network
Johannes Thiele, Paris (FR); and Olivier Bichler, Vieille Eglise en Yvelines (FR)
Assigned to COMMISSARIAT A L'ENERGIE ATOMIQUE ET AUX ENERGIES ALTERNATIVES, Paris (FR)
Appl. No. 17/287,277
Filed by COMMISSARIAT A L'ENERGIE ATOMIQUE ET AUX ENERGIES ALTERNATIVES, Paris (FR)
PCT Filed Oct. 22, 2019, PCT No. PCT/EP2019/078669
§ 371(c)(1), (2) Date Apr. 21, 2021,
PCT Pub. No. WO2020/083880, PCT Pub. Date Apr. 30, 2020.
Claims priority of application No. 1859760 (FR), filed on Oct. 23, 2018.
Prior Publication US 2021/0397968 A1, Dec. 23, 2021
Int. Cl. G06N 3/084 (2023.01); G06N 3/048 (2023.01); G06N 3/049 (2023.01); G06N 3/065 (2023.01)
CPC G06N 3/084 (2013.01) [G06N 3/048 (2023.01); G06N 3/049 (2013.01); G06N 3/065 (2023.01)] 23 Claims
OG exemplary drawing
 
1. A spiking artificial neuron (Ni,l), implemented by a calculating device, the neuron being intended to be integrated within a neural network comprising multiple successive layers (Cl−1, Cl, Cl+1), each layer being formed of multiple neurons, the neural network being configured so as to execute a training mechanism comprising a first phase of propagating data from an input layer to an output layer, and a second phase of back-propagating errors from the output layer to the input layer, the spiking artificial neuron (Ni,l) comprising, in order to execute the second back-propagation phase:
a first input/output interface (E/S1) capable of receiving binary or ternary error signals weighted by synaptic coefficients,
an error calculation module (ER_LOC) configured so as to calculate a binary or ternary local error signal (δi,l) from a binary or ternary intermediate signal (Zi,l) generated by the neuron in response to the received error signals and from an estimate (a′i,l) of the derivative of an equivalent activation function (FAE) implemented by the neuron (Ni,l) during the first data propagation phase,
a second input/output interface (E/S2) capable of propagating the binary or ternary local error signal (δi,l) to multiple synapses in the form of spikes.