Login| Sign Up| Help| Contact|

Patent Searching and Data


Title:
NEURAL NETWORK ARCHITECTURE SEARCH METHOD AND APPARATUS
Document Type and Number:
WIPO Patent Application WO/2022/037039
Kind Code:
A1
Abstract:
Disclosed are a neural network architecture search method and apparatus. The specific implementation scheme is as follows: the method comprises: acquiring configuration information of a search task, and generating, according to the configuration information, multiple candidate model structures and corresponding parameters; creating multiple training tasks according to the multiple candidate model structures and the parameters; scheduling the multiple training tasks to corresponding physical nodes, such that the respective physical nodes execute corresponding training tasks in parallel, and performing computation to obtain training results of the multiple training tasks; and if the training results meet a preset training condition in the configuration information, obtaining a search result corresponding to the search task. The invention not only improves search efficiency, but also increases a utilization rate of cluster resources, thereby enhancing scalability of hardware resources being used and hardware acceleration resources, and meeting the requirement of large-scale searching.

Inventors:
QIAO XIAOYA (CN)
LIU GUOBAO (CN)
ZHOU YONGKAI (CN)
Application Number:
PCT/CN2021/080497
Publication Date:
February 24, 2022
Filing Date:
March 12, 2021
Export Citation:
Click for automatic bibliography generation   Help
Assignee:
CHINA UNIONPAY CO LTD (CN)
International Classes:
G06F9/48; G06F9/455; G06F9/50; G06F16/953; G06N3/04; G06N3/08
Foreign References:
CN112000450A2020-11-27
CN111325356A2020-06-23
CN111324630A2020-06-23
CN111063000A2020-04-24
CN110543944A2019-12-06
US9390370B22016-07-12
Attorney, Agent or Firm:
ZHONG LUN LAW FIRM (CN)
Download PDF: