Multi-view Clustering via Matrix Decomposition and Partition Alignment
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Current multi-view clustering methods, such as those using Non-negative Matrix Factorization (NMF), face limitations in fully utilizing original data for discrimination, sharing information between views, and improving fusion strategies, leading to insufficient representation learning and potential noise in early fusion.
Innovation Solution
A multi-view clustering method based on matrix decomposition and multi-partition alignment, which involves acquiring and decomposing multi-view data to obtain basic partition matrices, fusing and aligning them using column transformation, constructing an objective function, and optimizing it through alternating optimization to achieve a unified partition matrix for spectral clustering.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Loss of information
If early fusion is used to fuse feature representations or graph structures of multiple views into a common representation, then the clustering algorithm can integrate information from different views, but the complexity of the algorithm increases and the computational cost rises
Solution Approach 1:
The patent segments the fusion process into two distinct stages: early fusion of feature representations and late fusion of clustering results. This segmentation allows the algorithm to handle information fusion at multiple levels, reducing the complexity burden on any single fusion operation while maintaining comprehensive information integration across views.
Solution Approach 2:
The patent introduces graph structures as intermediary representations that mediate between raw feature representations and final clustering results. These graph structures serve as a bridge that simplifies the fusion process by capturing relationships in a structured format that is easier to integrate across multiple views compared to raw high-dimensional features.
2Reliability
If multiple clustering results are obtained by spectral embedding for each view and used to influence original representations of other views in collaborative training, then the clustering performance improves, but the time cost and computational complexity increase
Solution Approach 1:
The patent performs spectral embedding and obtains clustering results for each view in advance before the main collaborative training process. These preliminary clustering results are then used as guidance to influence the original representations, avoiding the need for iterative recalculations during training and thus reducing the overall computational time while maintaining performance improvements.
Solution Approach 2:
The patent applies collaborative training influence only to the extent necessary - using the preliminary clustering results to guide the representation learning process without requiring full re-optimization of all parameters. This partial action approach achieves performance improvement while avoiding the excessive computational cost of complete re-training.
3Loss of information
If deep semi-NMF model is used to explore complex hierarchical information with implicit low-level latent attributes, then the information extraction capability improves, but the model complexity and difficulty of optimization increase
Solution Approach 1:
The patent segments the deep semi-NMF model into multiple hierarchical layers, each responsible for extracting specific levels of information. This segmentation allows the complex information extraction task to be divided into manageable stages, where each layer processes and transforms data in a controlled manner, reducing the optimization difficulty compared to a monolithic deep model.
Solution Approach 2:
The patent applies preliminary constraints and initialization strategies to the deep semi-NMF model before the main optimization process. By pre-configuring the model structure and initial parameters based on the specific characteristics of multi-view data, the optimization process becomes more stable and converges faster, reducing the practical difficulty of training complex hierarchical models.
Data Source
AI summary
A multi-view clustering method and system based on matrix decomposition and multi-partition alignment are provided. The multi-view clustering method based on matrix decomposition and multi-partition alignment includes: S1: acquiring a clustering task and a target data sample; S2: decomposing multi-view data corresponding to the acquired clustering task and the acquired target data sample through a multi-layer matrix to obtain a basic partition matrix of each view; S3: fusing and aligning the obtained basic partition matrix of each view by using column transformation to obtain a consistent fused partition matrix; S4: unifying the obtained basic partition matrix of each view and the consistent fused partition matrix, and constructing an objective function corresponding to the unified partition matrix; S5: optimizing the constructed objective function to obtain an optimized unified partition matrix; and S6: performing spectral clustering on the obtained optimized unified partition matrix to obtain a final clustering result.