K-Means Clustering with Distribution Boundaries for Faster Assignment
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing k-means clustering algorithms are inefficient and resource-intensive due to their reliance on single-value datapoints, which do not account for the inherent flexibility and variability in data point values, leading to suboptimal clustering results.
Innovation Solution
The algorithm modifies datapoints to include distribution boundaries, calculating effective distances between these boundaries and centroids, and assigns datapoints to clusters based on these distances, allowing for recomputation of centroids to improve clustering accuracy and reduce resource usage.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If traditional k-means clustering uses single-value data points, then the algorithm is simple to implement, but the clustering accuracy is suboptimal due to ignoring data variability
Solution Approach 1:
The patent transforms the data representation from single-value points to distributions characterized by mean and variance parameters. This parameter change allows the algorithm to account for data variability while maintaining a manageable computational framework through modified distance calculations that incorporate both mean and variance components.
Solution Approach 2:
The patent adds a new dimension to the data representation by incorporating variance as an additional parameter alongside the mean. This dimensional expansion from 1D (single value) to 2D (mean, variance) enables the clustering algorithm to capture data uncertainty and variability, improving clustering accuracy without excessive complexity.
2Productivity
If k-means clustering iteratively assigns data points to nearest centroids, then clusters are formed based on minimum distance, but computation time increases with large data sets
Solution Approach 1:
The patent extracts and precomputes the variance component from the distance calculation, separating it from the mean-based distance computation. This extraction allows the algorithm to reuse variance information across multiple iterations and data points, reducing redundant computations and improving efficiency on large data sets.
Solution Approach 2:
The patent performs preliminary computation of distribution parameters (mean and variance) before the main clustering iterations. By precomputing these statistical characteristics, the algorithm reduces the computational burden during iterative assignments, thereby decreasing overall computation time while maintaining clustering quality.
3Adaptability or versatility
If k-means clustering uses fixed centroids, then the algorithm converges quickly, but it fails to account for the statistical scope and flexibility of data points
Solution Approach 1:
The patent introduces dynamics to the centroid representation by transforming fixed centroids into distribution-based centroids with both mean and variance parameters. This dynamic representation allows centroids to adapt to the statistical scope of data points, accommodating data flexibility while maintaining reasonable convergence through the structured parameter updates.
4Reliability
If traditional clustering ignores data distribution boundaries, then the algorithm is computationally efficient, but clustering results do not reflect inherent data variability
Solution Approach 1:
The patent changes the computational parameters from simple distance metrics to distribution-aware metrics that incorporate variance. This parameter change enables the algorithm to reflect inherent data variability in clustering results, improving reliability while the efficient variance utilization keeps energy consumption manageable.
Data Source
AI summary
In a computer-implemented method for improving a k-means clustering algorithm, a processor modifies a first data point of a data set to include a first distribution boundary. A processor calculates a first effective distance between the first distribution boundary and a first centroid of a cluster. A processor assigns the first data point to the first cluster. A processor recomputes the first centroid based on the first data point and the first distribution boundary.