Title:
NEURAL NETWORK MODEL TRAINING METHOD AND APPARATUS, DEVICE, AND STORAGE MEDIUM
Document Type and Number:
WIPO Patent Application WO/2022/222649
Kind Code:
A1
Abstract:
A neural network model training method and apparatus, a device, and a storage medium, relating to the field of artificial intelligence. The method comprises: generating a corresponding first weight parameter for a network layer in a neural network model (201); performing quantization loss simulation on the first weight parameter on the basis of the number of quantization bits to obtain a second weight parameter (202); inputting a training sample into the neural network model using the second weight parameter to obtain a sample inference result (203); and determining an inference loss of the neural network model on the basis of the sample inference result, and adjusting the first weight parameter on the basis of the inference loss (204). By using the solution of embodiments of the present application, the quantization loss in subsequently quantizing the trained neural network model can be reduced, and the precision of the quantized neural network model can be improved.
More Like This:
Inventors:
ZHAO JUANPING (CN)
Application Number:
PCT/CN2022/081023
Publication Date:
October 27, 2022
Filing Date:
March 15, 2022
Export Citation:
Assignee:
GUANGDONG OPPO MOBILE TELECOMMUNICATIONS CORP LTD (CN)
International Classes:
G06N3/08
Foreign References:
| CN110929837A | 2020-03-27 | |||
| CN110969251A | 2020-04-07 | |||
| CN110555508A | 2019-12-10 | |||
| CN112101524A | 2020-12-18 | |||
| CN111783957A | 2020-10-16 | |||
| US20190347550A1 | 2019-11-14 |
Attorney, Agent or Firm:
BEIJING SAN GAO YONG XIN INTELLECTUAL PROPERTY AGENCY CO., LTD. (CN)
Download PDF: