南方科技大学知识苑(SUSTech KC): 神经网络的压缩方法和装置、设备、介质

Title	神经网络的压缩方法和装置、设备、介质
Alternative Title	Neural network compression method and device, equipment and medium
Author	洪文静; 李皈颖; 杨鹏 ; 唐珂
First Inventor	洪文静
Original applicant	南方科技大学
First applicant	南方科技大学
Address of First applicant	518055 广东省深圳市南山区西丽学苑大道1088号
Current applicant	南方科技大学
Address of Current applicant	518055 广东省深圳市南山区西丽学苑大道1088号 (广东,深圳,南山区)
First Current Applicant	南方科技大学
Address of First Current Applicant	518055 广东省深圳市南山区西丽学苑大道1088号 (广东,深圳,南山区)
Application Number	CN202111311435.4
Application Date	2021-11-08
Open (Notice) Number	CN114154634A
Date Available	2022-03-08
Status of Patent	实质审查
Legal Date	2022-03-25
Subtype	发明申请
SUSTech Authorship	First
Abstract	本实施例提供一种神经网络的压缩方法和装置、设备、介质，属于计算机应用技术领域。该方法包括：获取在硬件上运行的预设神经网络；对所述预设神经网络进行压缩处理，以生成至少两个第一压缩神经网络；获取每一所述第一压缩神经网络对应于所述硬件的硬件性能指标；计算所述预设神经网络的第一准确率和每一所述第一压缩神经网络的第二准确率；根据所述硬件性能指标、所述第一准确率、多个所述第二准确率对多个所述第一压缩神经网络进行初步筛选，得到初始神经网络集；根据预设条件对所述初始神经网络集进行二次筛选，得到目标压缩神经网络。根据硬件性能指标和准确率来指导神经网络的压缩，能够实现神经网络在多种硬件上的快速部署。
Other Abstract	The embodiment of the invention provides a neural network compression method and device, equipment and a medium, and belongs to the technical field of computer application. The method comprises the following steps: acquiring a preset neural network running on hardware; performing compression processing on the preset neural network to generate at least two first compression neural networks; obtaining a hardware performance index, corresponding to the hardware, of each first compression neural network; calculating a first accuracy rate of the preset neural network and a second accuracy rate of each first compression neural network; according to the hardware performance index, the first accuracy rate and the plurality of second accuracy rates, carrying out preliminary screening on the plurality of first compression neural networks to obtain an initial neural network set; and performing secondary screening on the initial neural network set according to a preset condition to obtain a target compressed neural network. Compression of the neural network is guided according to the hardware performance index and the accuracy, and rapid deployment of the neural network on various kinds of hardware can be achieved.
CPC Classification Number	G06N3/082
IPC Classification Number	G06N3/08
INPADOC Legal Status	（ENTRY INTO FORCE OF REQUEST FOR SUBSTANTIVE EXAMINATION）[2022-03-25][CN]
INPADOC Patent Family Count	1
Extended Patent Family Count	1
Priority date	2021-11-08
Patent Agent	洪铭福
Agency	广州嘉权专利商标事务所有限公司
URL	[Source Record]
Data Source	PatSnap
Document Type	Patent
Identifier	http://kc.sustech.edu.cn/handle/2SGJ60CL/532889
Department	Department of Computer Science and Engineering 工学院_斯发基斯可信自主研究院
Recommended Citation GB/T 7714	洪文静,李皈颖,杨鹏,等. 神经网络的压缩方法和装置、设备、介质.

Files in This Item:
There are no files associated with this item.

If you have any objections to this item, please fill out the form below and the administrator will contact you as soon as possible.
Content:
Email：	*
Affiliation Sequential No:
Verification Code:	Refresh

Any comments and suggestions are welcomed.
Title:	*
Content:
Email：	*
Verification Code:	Refresh