SEARCH METHOD FOR MACHINE LEARNING MODEL AND RELATED APPARATUS AND DEVICE

Title:

SEARCH METHOD FOR MACHINE LEARNING MODEL AND RELATED APPARATUS AND DEVICE

Document Type and Number:

WIPO Patent Application WO/2021/135707

Kind Code:

A1

Abstract:

Embodiments of the present invention specifically relate to the technical field of artificial intelligence, and disclosed therein are a search method for machine learning model and a related apparatus and device. The method comprises: before searching and quantizing a model, generating multiple pure bit models according to a model to be quantized, acquiring the evaluation parameters of each layer structure in the multiple pure bit models, selecting a candidate model from a candidate set for training and testing, and once a target model is obtained, it is possible to determine the quantization weight of each layer structure in the target model according to the network structure of the target model and the evaluation parameters of each layer structure in the target model, so as to quantize the layer structure of which the quantization weight is the biggest in the target model and add the quantized model into the candidate set, which can reduce frequent information exchange with a terminal and raise the efficiency of model search and model quantization.

Inventors:

YU QINGHUA (CN)
LIU MOHAN (CN)
SUI ZHICHENG (CN)
ZHOU LI (CN)
BAI LIXUN (CN)

Application Number:

PCT/CN2020/130043

Publication Date:

July 08, 2021

Filing Date:

November 19, 2020

Export Citation:

Click for automatic bibliography generation Help

Assignee:

HUAWEI TECH CO LTD (CN)

International Classes:

G06N5/04; G06N20/00

Foreign References:

CN111178546A	2020-05-19
CN110555508A	2019-12-10
CN108734266A	2018-11-02
CN109190754A	2019-01-11
US20190347550A1	2019-11-14
CN201911419960A	2019-12-31

Other References: