EP 3871153 A1 20210901 - BACKPROPAGATION OF ERRORS IN PULSED FORM IN A PULSED NEURAL NETWORK
Title (en)
BACKPROPAGATION OF ERRORS IN PULSED FORM IN A PULSED NEURAL NETWORK
Title (de)
RÜCKAUSBREITUNG VON FEHLERN IN GEPULSTER FORM IN EINEM GEPULSTEN NEURONALEN NETZWERK
Title (fr)
RETRO-PROPAGATION D'ERREURS SOUS FORME IMPULSIONNELLE DANS UN RESEAU DE NEURONES IMPULSIONNELS
Publication
Application
Priority
- FR 1859760 A 20181023
- EP 2019078669 W 20191022
Abstract (en)
[origin: WO2020083880A1] The invention proposes a novel implementation of an error backpropagation algorithm that is suited to the hardware constraints of a device implementing a pulsed neural network. In particular, the invention uses binary or ternary coding of the errors calculated during the backpropagation phase to adapt its implementation to the constraints of the network and thus avoid the use of floating-point multiplication operators. More generally, the invention proposes an overall adaptation of the backpropagation algorithm to the specific constraints of a pulsed neural network. In particular, the invention allows the same propagation infrastructure to be used for data propagation and for error backpropagation during the learning phase. The invention proposes a generic implementation of a pulsed neuron that is suitable for implementing any type of pulsed neural network, in particular convolutional networks.
IPC 8 full level
G06N 3/08 (2006.01); G06N 3/04 (2006.01); G06N 3/063 (2006.01)
CPC (source: EP US)
G06N 3/048 (2023.01 - EP US); G06N 3/049 (2013.01 - EP US); G06N 3/065 (2023.01 - US); G06N 3/084 (2013.01 - EP US); G06N 3/045 (2023.01 - EP); G06N 3/065 (2023.01 - EP)
Designated contracting state (EPC)
AL AT BE BG CH CY CZ DE DK EE ES FI FR GB GR HR HU IE IS IT LI LT LU LV MC MK MT NL NO PL PT RO RS SE SI SK SM TR
Designated extension state (EPC)
BA ME
DOCDB simple family (publication)
FR 3087560 A1 20200424; EP 3871153 A1 20210901; JP 2022504942 A 20220113; JP 7433307 B2 20240219; US 2021397968 A1 20211223; WO 2020083880 A1 20200430
DOCDB simple family (application)
FR 1859760 A 20181023; EP 19787001 A 20191022; EP 2019078669 W 20191022; JP 2021520563 A 20191022; US 201917287277 A 20191022