Title:
NEURAL NETWORK-BASED ADJUSTMENT METHOD, APPARATUS AND DEVICE
Document Type and Number:
WIPO Patent Application WO/2019/091443
Kind Code:
A1
Abstract:
A neural network-based adjustment method, apparatus and device. The method comprises: obtaining a neural network to be adjusted and input data (S101); inputting the neural network to be adjusted, into an inference engine (S102); adjusting the neural network to be adjusted for multiple times by using a component of the inference engine, so as to obtain a neural network adjusted each time (S103); for the neural network adjusted each time, inputting the input data into the adjusted neural network, so as to obtain output data this time (S104); and comparing the output data obtained each time, and determining an adjusted neural network corresponding to the output data at an optimal time (S105). The adjusted neural network corresponding to the output data at an optimal time is a neural network that is most adaptive to the input data. The method improves the adaptivity of the neural network to data.
Inventors:
YE TINGQUN (CN)
Application Number:
PCT/CN2018/114693
Publication Date:
May 16, 2019
Filing Date:
November 09, 2018
Export Citation:
Assignee:
HANGZHOU HIKVISION DIGITAL TEC (CN)
International Classes:
G06N3/06
Foreign References:
| CN103544705A | 2014-01-29 | |||
| CN106408001A | 2017-02-15 | |||
| CN106599883A | 2017-04-26 | |||
| CN106227851A | 2016-12-14 |
Attorney, Agent or Firm:
PATENTSINO IP FIRM (CN)
Download PDF: