Ising Machine Hierarchical Clustering Division Candidate Storage

Resolve Bottlenecks,
Find Innovative Solutions
Generate Solutions

Solution Overview

Problem

The execution time of divisive hierarchical clustering increases with the number of nodes, as the number of division patterns grows, leading to longer computation times for the Ising machine in determining optimal cluster divisions.

Innovation Solution

A data processing system that uses an Ising machine to calculate division candidates for cluster divisions based on an Ising-type evaluation function, with an information processing device selecting candidates and storing unselected ones for subsequent cluster divisions, reducing the number of computations required.

Engineering Contradictions & Design Principles

VSEngineering Contradiction Analysis

1Measurement precision

If the Ising machine calculates division candidates for each division pattern of the number of nodes of each cluster, then the accuracy of cluster division is improved, but the execution time of hierarchical clustering becomes longer

Engineering Contradiction:
Improveaccuracy of cluster divisionVSAvoidexecution time of hierarchical clustering
Core Design Contradiction:
Measurement precisionVSLoss of time

Solution Approach 1:

The patent applies preliminary action by pre-calculating and storing division candidate information for clusters that have been divided in previous steps. When a cluster is divided, the division candidate information is stored in a storage unit, and this stored information is reused for subsequent divisions of the resulting sub-clusters, avoiding redundant calculations and reducing execution time

Inventive Principle:
Principle #10Preliminary action

Solution Approach 2:

The patent implements universality by making the division candidate information serve multiple purposes. The same division candidate information stored from a parent cluster division is reused for multiple child cluster divisions, allowing one calculation result to benefit multiple subsequent operations across different cluster hierarchies

Inventive Principle:
Principle #6Universality (Multi-functionality)

Data Source

PatentEP4528600A1Data processing system, program, and data processing method
Publication Date: 2025.03.26 FUJITSU LTD
  • EP4528600A1 patent drawingFigure 1
  • EP4528600A1 patent drawingFigure 2
  • EP4528600A1 patent drawingFigure 3

AI summary

A data processing system includes: an Ising machine that calculates a first division candidate based on an Ising-type evaluation function; and an information processing device configured to obtain, from the Ising machine, division candidate information that represents the first division candidates for individual first division patterns; select one of the first division candidates and divide the first cluster into a second and a third clusters; store the division candidate information of the first division candidate that has not been selected; determine whether or not the unselected first division candidate that corresponds to a second division pattern of the second cluster exists based on the stored division candidate information; and when the unselected first division candidate is determined to exist, select one of second division candidates of the second cluster and divide the second cluster into a fourth and a fifth clusters.