Method and system for fast updating of machine learning model network in dynamic environment

By combining parameter importance detection and selective updating with regularization and iterative optimization, the problems of slow update speed and high resource consumption of machine learning models in dynamic environments are solved, achieving fast and stable model updates, which are suitable for real-time application scenarios such as smart cities and autonomous driving.

CN122152333APending Publication Date: 2026-06-05浪潮智慧城市科技有限公司
View PDF 0 Cites 0 Cited by

Patent Information

Authority / Receiving Office
CN · China
Patent Type
Applications(China)
Current Assignee / Owner
浪潮智慧城市科技有限公司
Filing Date
2026-01-13
Publication Date
2026-06-05

AI Technical Summary

Technical Problem

In dynamic environments, existing machine learning model update strategies consume huge amounts of computing resources and have long update times, making it difficult to meet the needs of real-time applications. Furthermore, incremental learning methods struggle to balance learning efficiency and stability and cannot accurately locate parameters that are sensitive to new data.

Method used

By detecting the importance of parameters, a subset of key parameters is located and selectively updated. Combined with regularization and iterative optimization, only parameters that are sensitive to new data are updated, while the remaining parameters are frozen, reducing the computational and storage burden and maintaining model stability.

Benefits of technology

It enables fast and efficient model updates in dynamic environments, improves update speed, reduces computing resource requirements, is suitable for resource-constrained edge devices, and maintains model stability and generalization ability.

✦ Generated by Eureka AI based on patent content.

Smart Images

  • Figure CN122152333A_ABST
    Figure CN122152333A_ABST
Patent Text Reader

Abstract

The application discloses a kind of dynamic environment in machine learning model network fast updating method and system, belong to machine learning and wisdom city technical field, based on parameter importance detection realizes network update, including: parameter importance detection: for pre-training network and new data set, design influence function based on gradient norm, evaluate the importance of each parameter to new data, locate key parameter subset;Selective parameter update: after determining the parameter subset to be updated, selective parameter update is carried out, only update key parameters, freeze the rest of parameters;Regularization and iterative optimization: by adaptive regularization mechanism, the variation amplitude of key parameters is constrained to prevent excessive deviation from original state;Through iterative monitoring and periodic fine-tuning, maintain model stability and generalization ability.The application can quickly detect parameter importance, accurately locate update target, while maintaining model stability, realize dynamic update.
Need to check novelty before this filing date? Find Prior Art