Quantum Circuit Training With Fine-Tuned Models for Faster Classification

Resolve Bottlenecks,
Find Innovative Solutions
Generate Solutions

Solution Overview

Problem

Existing quantum circuits face challenges in improving classification accuracy due to decreased relevance between training datasets and increased training time, particularly when transitioning from classical to quantum models.

Innovation Solution

A method involving a trained model that is updated using a specific dataset to enhance classification accuracy, followed by training a quantum circuit based on the updated model to output classification results aligned with the feature amounts extracted from the data.

Engineering Contradictions & Design Principles

VSEngineering Contradiction Analysis

1Measurement precision

If a trained model is updated based on a dataset with low relevance to improve classification accuracy of specific data type, then classification accuracy of the quantum circuit improves, but training time increases

Engineering Contradiction:
Improveclassification accuracyVSAvoidtraining time
Core Design Contradiction:
Measurement precisionVSLoss of time

Solution Approach 1:

The trained model is updated in advance based on the specific type of data before training the quantum circuit. This preliminary update ensures that when the quantum circuit training begins, the model is already optimized for the target data distribution, reducing the training time required while maintaining high classification accuracy.

Inventive Principle:
Principle #10Preliminary action

Solution Approach 2:

Instead of completely retraining the model from scratch on the specific dataset, only partial updates are applied to the existing trained model. This selective updating approach reduces training time while still improving classification accuracy for the specific data type, avoiding the excessive time cost of full retraining.

Inventive Principle:
Principle #16Partial or excessive action

2Reliability

If the trained model is updated using a dataset with low relevance to prevent overlearning, then generalization ability improves, but classification accuracy on specific data type may decrease

Engineering Contradiction:
Improvegeneralization abilityVSAvoidclassification accuracy
Core Design Contradiction:
ReliabilityVSMeasurement precision

Solution Approach 1:

The model update process applies different treatment to different parts of the model architecture. Specifically, certain layers or parameters are updated based on the specific type of data while other parts maintain their generalization capabilities. This localized updating approach allows the model to specialize for the target data type without sacrificing overall generalization ability.

Inventive Principle:
Principle #3Local quality

Solution Approach 2:

The update process modifies specific parameters of the trained model based on the specific type of data, while keeping other parameters unchanged. This selective parameter adjustment allows the model to adapt to the target data distribution for improved classification accuracy while maintaining the generalization properties encoded in the unchanged parameters.

Inventive Principle:
Principle #35Parameter changes

Data Source

PatentEP4625257A1Information processing program, information processing method, and information processing apparatus
Publication Date: 2025.10.01 FUJITSU LTD
  • EP4625257A1 patent drawingFigure 1
  • EP4625257A1 patent drawingFigure 2
  • EP4625257A1 patent drawingFigure 3

AI summary

An information processing program causing a computer to execute processing including: acquiring a trained model that has a function of outputting a classification result of input data in accordance with a feature amount extracted from the data; updating the acquired trained model based on a dataset that includes specific type of data such that classification accuracy of the specific type of data is improved; and controlling an arithmetic unit to train, based on the dataset, a quantum circuit that has a function of outputting a classification result of the specific type of data in accordance with a feature amount extracted from the specific type of data by the updated trained model.