A label distribution learning method and system based on adaptive particle swarm clustering

CN122114231APending Publication Date: 2026-05-29EAST CHINA JIAOTONG UNIVERSITY
View PDF 0 Cites 0 Cited by

Patent Information

Authority / Receiving Office
CN · China
Patent Type
Applications(China)
Current Assignee / Owner
EAST CHINA JIAOTONG UNIVERSITY
Filing Date
2026-04-22
Publication Date
2026-05-29

AI Technical Summary

Technical Problem

Existing label distribution learning methods are prone to failure under high-dimensional or non-spherical distributions, resulting in unstable partitioning and susceptibility to noise and human interference, leading to inaccurate partitioning.

Method used

An adaptive particle-sphere clustering method is adopted. By dynamically defining the cluster number search interval, combining the particle-sphere clustering algorithm and the maximum entropy model, the optimal clustering results are selected using the silhouette coefficient, a label distribution learning model is constructed, and the optimal parameters are solved iteratively through gradient descent.

Benefits of technology

It improves the accuracy of label distribution learning, reduces computational costs, avoids subjective bias caused by manually specifying the number of clusters, and enhances the modeling effect of local correlation.

✦ Generated by Eureka AI based on patent content.

Smart Images

  • Figure CN122114231A_ABST
    Figure CN122114231A_ABST
Patent Text Reader

Abstract

The application discloses a kind of based on self-adapting granular ball clustering's mark distribution learning method and system, it is related to machine learning technical field, this method includes: obtaining the mark distribution dataset needing to carry out mark distribution learning;Combining the sample scale of training set, dynamically delimiting cluster number search interval, then in cluster number search interval to candidate cluster number is one by one traversed, clustering is carried out to dataset, obtains clustering result;Select the clustering result that the outline coefficient meets set condition as final clustering result, based on final clustering result, obtain the cluster label y of each sample, cluster prototype matrix P and sample-cluster indication matrix U;Based on maximum entropy model, construct mark distribution learning model, obtain target mark distribution learning model;Based on target mark distribution learning model, predict the mark distribution of test set.The application solves the problem that in the mark distribution learning process in prior art, inaccurate and susceptible to noise interference and human factor interference is divided.
Need to check novelty before this filing date? Find Prior Art