Frequency-Based Centroid Initialization for Memory Read Clustering
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
The selection of initial centroids in k-means clustering (KMC) algorithms significantly affects performance in terms of running time and accuracy, leading to inefficiencies in data clustering and read voltage determination for memory devices.
Innovation Solution
A data clustering system that determines initial centroids based on the frequency of data values, reducing the number of centroids needed and optimizing the grouping process, and a read voltage determination apparatus that uses this system to generate a read retry table for memory devices.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If initial centroids are selected using conventional methods in k-means clustering, then the clustering process can proceed, but the running time increases and accuracy decreases due to suboptimal centroid selection
Solution Approach 1:
The patent applies preliminary action by pre-processing the data to identify frequent data values before the main clustering process. The initialization circuit counts frequencies of data values in advance and uses this information to select initial centroids, avoiding the need for random selection or iterative adjustments during the clustering process itself.
Solution Approach 2:
The patent changes the parameter selection criterion from random or arbitrary initial centroid selection to frequency-based selection. By using the frequency of data values as the selection criterion, the system identifies centroids that are more representative of actual data distributions, thereby improving clustering accuracy and reducing iterations required.
2Measurement precision
If the number of initial centroids is increased to improve clustering coverage, then clustering accuracy may improve, but the complexity of distance calculations and processing increases
Solution Approach 1:
The patent optimizes the number of initial centroids by using frequency analysis to identify the most representative data values. Instead of using a fixed or large number of centroids, the system determines the optimal number based on data characteristics, reducing unnecessary processing complexity while maintaining clustering accuracy.
3Productivity
If conventional centroid selection methods are used, then the k-means algorithm can execute, but the number of iterations required for convergence increases, reducing productivity
Solution Approach 1:
The patent performs preliminary frequency analysis to initialize centroids in optimal positions before the main clustering iterations begin. This preliminary action ensures that the clustering process starts from a more favorable state, reducing the number of iterations needed to reach convergence and improving overall processing speed.
Solution Approach 2:
The patent changes the initialization parameter selection from random or arbitrary values to frequency-based optimal values. This parameter change leads to faster convergence by reducing the distance between initial centroids and actual cluster centers, thereby decreasing the number of iterations required.
Data Source
AI summary
A data clustering system includes: an initialization circuit configured to receive a data set including a plurality of samples and a number K of groups, where K is a natural number, and determine at least a part number of initial centroids based on frequency of a data value of the sample, wherein the number of initial centroids corresponds to the number K of groups; and a classification circuit configured to group the plurality of samples based on a distance between the data value corresponding to the sample and each of centroids.