Iterative Vector Clustering for Faster High-Dimensional Convergence

Resolve Bottlenecks,
Find Innovative Solutions
Generate Solutions

Solution Overview

Problem

Existing clustering algorithms are sensitive to initial placement of cluster centers and suffer from the curse of dimensionality, making them unsuitable for time-constrained scenarios such as cybersecurity threat analysis and telecommunications network reconfiguration.

Innovation Solution

An iterative clustering method that reassesses the placement of vectors by reassigning outermost vectors to the nearest neighboring clusters, using fitness metrics like silhouette coefficient and between-cluster distance to refine the clustering solution.

Engineering Contradictions & Design Principles

VSEngineering Contradiction Analysis

1Measurement precision

If gradient descent is used to find maxima or minima in clustering algorithms, then the algorithms can optimize cluster placement, but they become highly sensitive to initial placement of cluster centers and slow to converge

Engineering Contradiction:
Improveclustering accuracyVSAvoidconvergence time
Core Design Contradiction:
Measurement precisionVSLoss of time

Solution Approach 1:

The patent applies preliminary action by using a space-partitioning data structure (such as a k-d tree or ball tree) to pre-organize the data points before clustering begins. This preprocessing step creates a hierarchical structure that enables efficient nearest-neighbor searches during the clustering process, eliminating the need for slow gradient descent iterations while maintaining high clustering accuracy through intelligent initial cluster center placement.

Inventive Principle:
Principle #10Preliminary action

2Adaptability or versatility

If traditional clustering algorithms are used, then they can group data into categories, but they suffer from the curse of dimensionality and require significant computational resources

Engineering Contradiction:
Improveclustering capabilityVSAvoidcomputational complexity
Core Design Contradiction:
Adaptability or versatilityVSDevice complexity

Solution Approach 1:

The patent introduces an intermediary space-partitioning data structure that acts as a mediator between the high-dimensional data points and the clustering algorithm. This intermediate structure organizes data in a way that reduces the computational complexity of distance calculations and cluster assignments, making the clustering process scalable to high-dimensional datasets without sacrificing clustering capability.

Inventive Principle:
Principle #24Intermediary (Mediator)

3Measurement precision

If manual labeling is used to train machine learning models, then the models can be trained accurately, but it is time-consuming and resource-intensive

Engineering Contradiction:
Improvemodel training accuracyVSAvoidtraining preparation time
Core Design Contradiction:
Measurement precisionVSLoss of time

Solution Approach 1:

The patent applies self-service by implementing unsupervised clustering that automatically groups unlabeled data points into meaningful categories without requiring manual annotation. The space-partitioning approach enables the algorithm to self-organize the data structure and identify clusters independently, eliminating the time-consuming manual labeling process while still producing accurate cluster assignments that can be used for model training.

Inventive Principle:
Principle #25Self-service

Data Source

PatentEP4703919A1Iterative vector clustering algorithm
Publication Date: 2026.03.04 BRITISH TELECOM PLC
  • EP4703919A1 patent drawingFigure 1
  • EP4703919A1 patent drawingFigure 2~3
  • EP4703919A1 patent drawingFigure 4~5

AI summary

A computer-implemented method of iteratively clustering a plurality of vectors, the computer-implemented method comprising generating a candidate solution based on a current solution by reassigning an outermost vector of a selected cluster to a cluster of a plurality of vector clusters containing a vector of the plurality of vector clusters other than the selected cluster with the shortest distance from that outermost vector.