Parallel Data Clustering with Dynamic Group Evaluation
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Current data clustering methods require assumptions about the number of groups, which can lead to poor grouping due to incorrect selection, and high-dimensional data makes visualization challenging for humans.
Innovation Solution
Perform multiple clustering attempts on a data set using different group numbers in parallel on a multi-core server, evaluating results to determine the optimal group number and visualizing the clustering for better differentiation and similarity among groups.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If multiple clustering attempts are performed with different group numbers in parallel, then the accuracy of grouping improves, but the computational time and resource consumption increase
Solution Approach 1:
The patent performs multiple clustering attempts with different group numbers in parallel before final evaluation, preparing multiple candidate groupings simultaneously. This preliminary parallel computation allows the system to have multiple pre-computed clustering results ready for quick comparison and selection, improving final grouping accuracy without sequentially wasting time
Solution Approach 2:
The patent dynamically evaluates multiple clustering results with different group numbers and selectively presents the optimal grouping based on evaluation metrics. The system adapts the number of groups dynamically rather than fixing it beforehand, allowing the clustering structure to optimize itself based on the data characteristics and evaluation criteria
2Adaptability or versatility
If data objects exist in high dimensional space, then the data can have many characteristics, but visualization of the data objects becomes challenging
Solution Approach 1:
The patent applies dimensionality reduction techniques to transform high-dimensional data objects into lower-dimensional representations suitable for human visualization. By changing the dimensional space from high-dimensional to 2D or 3D while preserving essential clustering structure and relationships, the system maintains the versatility of multi-characteristic data analysis while making the results visually accessible and interpretable
Data Source
AI summary
This disclosure provides various techniques that may allow for clustering a set of data objects and creating a visualization of the clustered data. For example, a user may desire to cluster data objects in a dataset to form meaningful groupings of the data objects. This disclosure provides, in one example of data object clustering and visualization, that a plurality of groupings of the data objects may be determined in parallel. These groupings may be evaluated and a particular grouping selected based on the evaluating. Data useable to visualize the particular grouping may be generated.