Proxy Graph Multi-Kernel Clustering for Outlier Robustness
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing later fusion multi-kernel clustering algorithms are susceptible to the quality of basic divisions, which can be affected by outliers and noises, leading to unsatisfactory clustering results, and struggle with handling non-linear transformations in multi-kernel data.
Innovation Solution
A later-fusion multiple kernel clustering method based on proxy graph improvement, which initializes a proxy graph improvement matrix, combines kernel k-means clustering with graph improvement methods, and performs spectral clustering to optimize basic divisions and fuse kernel information effectively.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Device complexity
If later fusion clustering algorithms are used, then computational and storage complexities are reduced, but clustering performance deteriorates due to susceptibility to outliers and noises in basic divisions
Solution Approach 1:
The patent introduces a proxy graph as an intermediary structure between the basic divisions from multiple kernel matrices and the final clustering result. The proxy graph learns a unified similarity relationship that mediates the fusion process, making the algorithm less sensitive to outliers and noises in individual basic divisions while maintaining computational efficiency
Solution Approach 2:
The patent transforms the clustering problem by changing from directly fusing basic division results to learning a proxy graph representation first. This parameter transformation allows the algorithm to capture global data structure information before performing fusion, improving robustness against local outliers and noises
2Ease of operation
If basic divisions are simply taken as linear transformations, then the algorithm is easy to implement, but it cannot handle non-linear transformations in multi-kernel data effectively
Solution Approach 1:
The patent moves the fusion process from the original data space to a new dimension - the proxy graph space. By learning a proxy graph that represents similarity relationships in a transformed space, the algorithm can capture non-linear patterns in multi-kernel data while maintaining the simplicity of graph-based clustering operations
3Device complexity
If kernel k-means clustering is performed separately on each basis kernel, then the clustering process is simple, but the quality of basic divisions is poor due to separation from fusion process
Solution Approach 1:
The patent establishes a feedback mechanism where the proxy graph learning process iteratively refines the basic divisions from multiple kernel matrices. The learned proxy graph provides feedback information that guides the optimization of basic divisions, improving their quality through multiple iterations while maintaining the simplicity of the overall process
Data Source
AI summary
A later-fusion multiple kernel clustering machine learning method and system based on proxy graph improvement is provided. The method includes: S1. acquiring a clustering task and a target data sample; S2. initializing a proxy graph improvement matrix; S3. running k-means clustering and graph improvement on each view corresponding to the acquisition of the clustering task and the target data sample, and constructing an objective function by combining kernel k-means clustering and graph improvement methods; S4. cyclically solving the objective function constructed in step S3 so as to obtain a graph matrix, which is fused with basic kernel information; and S5. performing spectral clustering on the obtained graph matrix, so as to obtain a final clustering result. By means of the method, an optimized basic division not only has information of a single kernel, but can also obtain global information by means of a proxy graph.