Neural network architecture search method and device

By performing multiple training tasks in parallel on multiple machine clusters, the problems of low search efficiency and poor resource utilization in the deep learning neural network architecture in the prior art are solved, and more efficient search and more efficient resource utilization are achieved.

CN112000450BActive Publication Date: 2025-03-25CHINA UNIONPAY
1 Cites 0 Cited by

Patent Information

Application Number
CN202010829782.5
Authority / Receiving Office
CN · China
Patent Type
Patents(China)
Current Assignee / Owner
Filing Date
2020-08-18
Publication Date
2025-03-25
Estimated Expiration
2040-08-18

AI Technical Summary

Technical Problem

The existing deep learning neural network architecture search technology is difficult to perform search tasks in parallel on a cluster composed of multiple machines, resulting in low search efficiency and poor cluster resource utilization.

Method used

By obtaining the configuration information of the search task, multiple candidate model structures and their parameters are generated, multiple training tasks are created and scheduled to physical nodes, parallel execution is implemented, and training results are counted to obtain search results.

Benefits of technology

It improves the efficiency of neural network architecture search and the utilization rate of cluster resources, enhances the scalability of hardware resources and hardware accelerated resources, and meets the needs of large-scale searches.

✦ Generated by Eureka AI based on patent content.
Patent Text Reader

Abstract

The present application discloses a neural network architecture search method and device, and the specific implementation scheme is as follows: the method includes: obtaining the configuration information of the search task, generating multiple candidate model structures and corresponding parameters according to the configuration information; creating multiple training tasks according to the multiple candidate model structures and parameters; scheduling multiple training tasks to corresponding physical nodes so that each physical node executes the corresponding training task in parallel, and counting the training results of multiple training tasks; and obtaining the search result corresponding to the search task when the training result meets the preset training conditions in the configuration information. Not only the search efficiency is improved, but also the utilization rate of cluster resources is improved, and the scalability of the used hardware resources and hardware acceleration resources is enhanced to meet the needs of large-scale search.
Need to check novelty before this filing date? Find Prior Art