Deep clustering method using diffusion map guided
By employing a diffusion-guided deep clustering method, utilizing approximate nearest neighbor search and an autoencoder, combined with a Sinkhorn equalization module and a loss function, the problem of clustering complexity for large-scale high-dimensional data is solved, achieving efficient clustering results.
CN122241293APending Publication Date: 2026-06-19NANTONG UNIV
Patent Information
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- NANTONG UNIV
- Filing Date
- 2026-03-12
- Publication Date
- 2026-06-19
Smart Images
Figure CN122241293A_ABST
Abstract
This invention discloses a deep clustering method guided by diffusion mapping, which learns neighborhood relationships through an approximate nearest neighbor search method. The neighborhood relationships are transformed into diffusion embeddings for each sample via diffusion mapping, avoiding the maintenance of a neighborhood relationship graph while capturing key neighborhood relationships. This invention reduces the time complexity of constructing a neighborhood relationship graph by calculating the neighborhood relationship matrix using an approximate nearest neighbor method, and avoids maintaining the neighborhood relationship graph by transforming neighborhood relationships into diffusion embeddings.
Need to check novelty before this filing date? Find Prior Art