US 11,941,526 B2
Methods, electronic devices, and computer-readable media for training, and processing data through, a spiking neuron network
Zhenzhi Wu, Beijing (CN); Qikun Zhang, Beijing (CN); and Yaolong Zhu, Beijing (CN)
Assigned to LYNXI TECHNOLOGIES CO., LTD., Beijing (CN)
Appl. No. 18/004,013
Filed by LYNXI TECHNOLOGIES CO., LTD., Beijing (CN)
PCT Filed Dec. 28, 2021, PCT No. PCT/CN2021/141981
§ 371(c)(1), (2) Date Dec. 30, 2022,
PCT Pub. No. WO2022/148272, PCT Pub. Date Jul. 14, 2022.
Claims priority of application No. 202110018629.9 (CN), filed on Jan. 7, 2021.
Prior Publication US 2023/0196102 A1, Jun. 22, 2023
Int. Cl. G06N 3/08 (2023.01); G06N 3/049 (2023.01)
CPC G06N 3/08 (2013.01) [G06N 3/049 (2013.01)] 16 Claims
OG exemplary drawing
 
1. A method for training spiking neuron network, comprising:
inputting a spiking signal corresponding to a training sample into a spiking neuron network to be trained; and
training the spiking neuron network based on a back propagation rule, wherein a forward propagation stage of training comprises: adjusting a spiking firing threshold corresponding to a target layer in the spiking neuron network according to confidence corresponding to the target layer,
wherein the confidence corresponding to the target layer is associated with an expected spiking firing sparsity of the target layer, and
wherein a spiking firing sparsity is a ratio of the number of neurons firing spiking in the target layer to the number of all neurons in the target layer;
wherein before the adjusting a spiking firing threshold corresponding to a target layer in the spiking neuron network according to confidence corresponding to the target layer, the method further comprises:
calculating an activation parameter corresponding to the target layer according to a membrane potential corresponding to the target layer;
the adjusting a spiking firing threshold corresponding to a target layer in the spiking neuron network according to confidence corresponding to the target layer comprises: calculating a target adjustment threshold according to the activation parameter and the confidence, and adjusting the spiking firing threshold based on the target adjustment threshold.