Distance-based clustering using QUBO formulations.

Area of Science:

• Computer Science
• Optimization
• Quantum Computing

Background:

• Clustering groups data, with Ising machines solving distance-based problems via QUBO.
• Existing Ising machine methods struggle with unevenly distributed data due to fixed cluster sizes.

Purpose of the Study:

• To develop an improved clustering method for unevenly distributed data using Ising machines.
• To enhance the performance and efficiency of Ising machine-based clustering algorithms.

Main Methods:

• A hybrid algorithm combining an iterative process with Ising machine discrete optimization and general-purpose computer parameter calculation.
• Utilized a low-latency Ising machine with a field-programmable gate array (FPGA) for simulated bifurcation algorithm to minimize communication overhead.
• Applied the method to cluster 200 unevenly distributed data points.

Main Results:

• The proposed hybrid method achieved an 18% higher clustering score than the simple Ising machine method for uneven data.
• Reduced communication overhead to approximately 20% of total execution time.
• Demonstrated efficient discrete optimization with 2000 variables performed 100 times per iteration.

Conclusions:

• Hybrid algorithms integrating Ising machines offer an efficient solution for practical, complex optimization problems, particularly in data clustering.
• The developed method shows significant improvements for unevenly distributed datasets.
• Ising machines, when combined with hybrid approaches, are viable for real-world computational challenges.