Model generation method and apparatus, and computer device and storage medium

By automating the grouping and merging of the changes in weight parameters of the models to be merged, the problem of model discrepancies caused by manual selection of key weight parameters in existing technologies is solved, achieving efficient and accurate model generation and performance improvement.

WO2026114044A1PCT designated stage Publication Date: 2026-06-04CHINA TELECOM CLOUD TECH CO LTD
View PDF 5 Cites 0 Cited by

Patent Information

Application Number
PCT/CN2025/135989
Authority / Receiving Office
WO · WO
Patent Type
Applications
Current Assignee / Owner
Priority Date
2024-11-26
Filing Date
2025-11-19
Publication Date
2026-06-04

AI Technical Summary

Technical Problem

Existing technologies rely on manual selection of key weight parameters during model fusion, resulting in a lack of data support and poor interpretability. This leads to significant differences between the fused model and the target model, failing to meet the needs of practical applications.

Method used

By obtaining the change in weight parameters of the models to be merged, selecting the change in target weight parameters as the grouping threshold, grouping according to similarity, and merging weight parameters in the target grouping results to generate the target model.

Benefits of technology

It achieves automated model merging, improves the accuracy and efficiency of generation, enhances the overall performance and generalization ability of the target model, and improves prediction accuracy and model interpretability.

✦ Generated by Eureka AI based on patent content.

Smart Images

  • Figure CN2025135989_04062026_PF_FP_ABST
    Figure CN2025135989_04062026_PF_FP_ABST
Patent Text Reader

Abstract

The present application relates to a model generation method and apparatus, and a computer device and a storage medium. The method comprises: acquiring weight parameter variations of models to be merged; selecting a target weight parameter variation from among the weight parameter variations, respectively grouping the weight parameter variations by using the target weight parameter variation as a grouping threshold, and determining a target grouping result on the basis of the similarity between grouping results; and acquiring a target group from the target grouping result, and merging weight parameters corresponding to the weight parameter variations in the target group, so as to obtain a target model.
Need to check novelty before this filing date? Find Prior Art

Citation Information

Patent Citations

  • Model generation method and model generation device based on variable grouping and electronic equipment

    CN112116443A

  • Model processing method, data processing method and computing device

    CN117829103A

  • Model generation method and device, computer equipment and storage medium

    CN119578591A

  • Model-building optimization

    US20090083680A1

  • Systems and methods for node weighting and aggregation for federated learning techniques

    US20230186167A1