US 11,853,875 B2
Neural network apparatus and method
Bodo Ruckauer, Zurich (CH); and Shih-Chii Liu, Zurich (CH)
Assigned to Samsung Electronics Co., Ltd., Suwon-si (KR); and UNIVERSITAET ZUERICH, Zurich (CH)
Filed by SAMSUNG ELECTRONICS CO., LTD., Suwon-si (KR); and UNIVERSITAET ZUERICH, Zurich (CH)
Filed on Oct. 23, 2018, as Appl. No. 16/168,024.
Claims priority of provisional application 62/575,635, filed on Oct. 23, 2017.
Claims priority of application No. 10-2018-0087650 (KR), filed on Jul. 27, 2018.
Prior Publication US 2019/0122110 A1, Apr. 25, 2019
Int. Cl. G06N 3/08 (2023.01); G06N 3/04 (2023.01)
CPC G06N 3/08 (2013.01) [G06N 3/04 (2013.01)] 20 Claims
OG exemplary drawing
 
1. A processor-implemented neural network method, the method comprising:
acquiring connection weight of an analog neural network (ANN) node of a pre-trained ANN;
generating a spiking neural network (SNN) by determining a firing rate of an SNN node of the SNN, corresponding to the ANN node, based on the connection weight and information indicating a timing at which a first SNN node initially fires to generate a first spike such that other spikes subsequent to the first spike are unused in determining the firing rate of the SNN node; and
performing an inference operation including recognition, liveness, and/or classification associated with input data, based on an output from the SNN being generated,
wherein the determining of the firing rate comprises:
determining first timing information based on previous timing information indicating a timing at which a second SNN node of a previous layer connected to the first SNN node fires, a connection weight between the first SNN node and the second SNN node, a neuron potential based on the first spike of the first SNN node and an instantaneous rate of change in the neuron potential, and
wherein the first timing information of each node included in the SNN is determined based on a reciprocal of an activation of a respectively corresponding node in the ANN.