BRAIN MACHINE INTERFACE DECODING METHOD BASED ON SPIKING NEURAL NETWORK

Title:

BRAIN MACHINE INTERFACE DECODING METHOD BASED ON SPIKING NEURAL NETWORK

Document Type and Number:

WIPO Patent Application WO/2022/257329

Kind Code:

A1

Abstract:

A brain machine interface decoding method based on a spiking neural network. The method comprises: (1) constructing a liquid state machine model based on a spiking neural network, wherein the liquid state machine model is composed of an input layer, an intermediate layer and an output layer, a connection weight from the input layer to the intermediate layer is Whx, a recurrent connection weight inside the intermediate layer is Whh, and a read weight from the intermediate layer to the output layer is Wyh; and (2) inputting a pulse electroencephalogram signal, and training each weight by using the following policies: (2-1) performing unsupervised training on the connection weight Whx by means of STDP; (2-2) setting the recurrent connection weight Whh of the intermediate layer by means of a distance model plus random connection; and (2-3) performing supervised training on the connection weight Wyh by means of ridge regression, establishing the mapping between liquid state information R(t) of the intermediate layer and output motion information Y(t), and finally outputting a predicted motion trajectory. By using the present method, a model can be quickly trained within a relatively short time, and a motion trajectory of an arm is predicted in real time, thereby improving the efficiency and accuracy.

More Like This:

WO/2021/232589	INTENTION IDENTIFICATION METHOD, APPARATUS AND DEVICE BASED ON ATTENTION MECHANISM, AND STORAGE MEDIUM
WO/2022/198233	EFFICIENT COMPRESSION OF ACTIVATION FUNCTIONS
WO/2019/207042	TEST AND TRAINING DATA

Inventors:

QI YU (CN)
FANG TAO (CN)
PAN GANG (CN)
WANG YUEMING (CN)

Application Number:

PCT/CN2021/126579

Publication Date:

December 15, 2022

Filing Date:

October 27, 2021

Export Citation:

Click for automatic bibliography generation Help

Assignee:

UNIV ZHEJIANG (CN)

International Classes:

G06N3/04; G06N3/06; G06N3/08

Domestic Patent References:

WO2020115746A1

2020-06-11

Foreign References:

CN113298242A	2021-08-24
CN112784976A	2021-05-11
CN111783973A	2020-10-16
CN110751032A	2020-02-04
CN112405542A	2021-02-26

Attorney, Agent or Firm:

HANGZHOU TIANQIN INTELLECTUAL PROPERTY AGENCY CO., LTD. (CN)

Download PDF:

View/Download PDF PDF Help

Previous Patent: COBALT-NITROGEN CO-DOPED THREE-DIMENSIONAL STRUCTURED CARBON MATERIAL, PREPARATION METHOD THEREFOR, ...

Next Patent: INCREMENTAL SEGMENTATION PROCESSING METHOD AND APPARATUS, COMPUTER DEVICE, AND STORAGE MEDIUM