Multi-dimensional training method and device of support vector machine
By discretizing and kernel function mapping the training sample data of the support vector machine, and optimizing the objective function in combination with the gradient descent algorithm, the problems of complex operations and low classification accuracy in the existing technology are solved, and the classification and analysis capabilities of SVM are improved. .
Patent Information
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- GUANGDONG POWER GRID CO LTD
- Filing Date
- 2021-11-30
- Publication Date
- 2022-03-15
- Estimated Expiration
- Not applicable · inactive patent
AI Technical Summary
The existing support vector machine (SVM) training method requires users to manually adjust the relationship between data. The operation is complex and difficult to mine data relationships in different dimensions, resulting in low classification accuracy and unable to meet actual needs.
By discretizing the training sample data set, the classification contribution parameters of each attribute feature vector are calculated, and the kernel function is used to perform data mapping on these parameters, and the objective function is optimized with the gradient descent algorithm to train the support vector machine model.
The classification and analysis capabilities of the support vector machine are improved, the data gain weight is determined through kernel function mapping, and the linear separability of the mapped dimensional data is enhanced, thereby improving the classification accuracy.
Abstract
Citation Information
Cited By
Data classification method of quantum support vector machine based on variable component sub-kernel
CN117591947A
A data classification method based on quantum support vector machine with variational quantum kernel
CN117591947B
BIM acquisition data processing method and system, electronic equipment and storage medium
CN117743955A