Cluster Model Persistence Using Representative Subsets for Accurate Updates
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing clustering algorithms in databases randomly select subsets of records for updating clusters, leading to improper classification and redundant clusters, resulting in inefficiency and information loss.
Innovation Solution
Implement a method that identifies a representative subset of data records based on spatial distance using an approximate nearest neighbor (ANN) and Density-Based Spatial Clustering of Applications with Noise (DBSCAN) algorithms to accurately group records, minimizing redundant clusters and information loss.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Productivity
If random subset selection is used to update clusters, then update speed is improved, but classification accuracy deteriorates and information loss increases
Solution Approach 1:
The patent applies preliminary action by pre-computing and storing representative records (centroids) for each cluster before updates occur. When new records are added, the system compares them against these pre-stored centroids to determine cluster membership, avoiding the need to randomly sample and re-process entire clusters during updates. This maintains both speed and accuracy.
Solution Approach 2:
The patent creates simplified copies of cluster data in the form of centroids - condensed representations that capture the essential characteristics of each cluster. These centroids serve as proxies for the full cluster data, enabling fast comparison and classification without requiring access to or processing of all original records in the cluster.
2Use of energy by moving object
If random subset selection is used for cluster updates, then computational resources are conserved, but information loss increases from 90% to 20%
Solution Approach 1:
The system creates compact centroid representations that preserve the essential information of each cluster in a condensed form. Instead of randomly sampling records (which loses 90% of information), the centroid calculation maintains the mean position and characteristics of all records in a cluster, achieving both resource efficiency and information preservation.
Solution Approach 2:
The patent transforms the representation of cluster data from individual records to aggregated centroid parameters. This parameter transformation condenses multiple records into a single representative point with specific coordinates and characteristics, reducing data volume while maintaining the statistical essence of the cluster for comparison purposes.
3Measurement precision
If entire clusters are reprocessed for each update, then classification accuracy is maintained, but training time increases by 70%
Solution Approach 1:
The patent extracts only the essential representative records (centroids) from each cluster for the purpose of comparison and classification. Instead of reprocessing entire clusters during updates, the system extracts and uses only these condensed representations, dramatically reducing the computational workload while maintaining classification accuracy through the preserved essential characteristics.
Solution Approach 2:
The system performs the computationally intensive centroid calculation in advance, before updates occur. This preliminary computation creates ready-to-use representations that can be quickly compared against new records during updates, avoiding the need to reprocess entire clusters at update time and thus reducing training time by 70%.
4Productivity
If random subset selection is used, then cluster redundancy increases, but data processing speed is improved
Solution Approach 1:
The patent uses centroid copies as unique representatives for each cluster, eliminating redundant storage of multiple similar records. Each cluster is represented by a single condensed centroid that captures its essential characteristics, preventing the creation of redundant clusters while maintaining fast processing speeds through the use of these compact representations.
Data Source
AI summary
Systems and methods to identify a data cluster for an input data record based on a spatial distance to a representative subset of data records corresponding to the data cluster. The representative subset of data records may be identified based on a spatial distance to other records in the cluster.