MODEL TRAINING METHOD AND RELATED DEVICE

WIPO Patent Application WO/2024/093294

A1

Disclosed in embodiments of the present application are a model training method and a related device, for use in performing acceleration training on a target model. In the present application, since sampling is performed in the target model, to obtain a sub-model of the target model, and the number of feature transformation layers of the sub-model is less than the number of feature transformation layers of the target model, and/or, and the size of the weight matrix of at least one of the feature transformation layers in the sub-model is smaller than the size of the weight matrix of the corresponding feature transformation layer in the target model, compared with training of the target model, training of the sub-model can improve the training efficiency. The sub-model is augmented, and the augmented model is trained to obtain a trained augmented model, so that the performance of the trained model can be guaranteed. Therefore, the technical solution of the present application improves the training efficiency while guaranteeing the performance of the model.

WO/2023/177319	OPERATION OF A NEURAL NETWORK WITH CONDITIONED WEIGHTS
JP2023545199	Model training method, human body posture detection method, apparatus, device and storage medium
WO/2023/242543	METHODS AND SYSTEMS FOR DETERMINING CORRECTNESS OF MACHINE LEARNING MODEL OUTPUT

TANG YEHUI (CN)
DING NING (CN)
HAN KAI (CN)
XU CHAO (CN)
WANG YUNHE (CN)

PCT/CN2023/103347

May 10, 2024

June 28, 2023

Click for automatic bibliography generation  

HUAWEI TECH CO LTD (CN)

G06N3/0464

CN113570029A	2021-10-29
US20210073646A1	2021-03-11
CN109242028A	2019-01-18
CN112070207A	2020-12-11
US20220027434A1	2022-01-27

SHENPAT INTELLECTUAL PROPERTY AGENCY (CN)

View/Download PDF      PDF Help

Previous Patent: METHOD FOR CARRYING OUT STENCIL COMPUTATIONS, AND APPARATUS

Next Patent: ANTI-SPARKING METHOD FOR ELECTRIC TOOL, AND CONTROL DEVICE AND COMPUTER-READABLE MEDIUM