Quantum Circuit Training Time Prediction for Architecture Selection

Resolve Bottlenecks,
Find Innovative SolutionsGenerate Solutions

Solution Overview

Problem

Classical machine learning techniques are inadequate for designing efficient quantum circuit architectures, and the dependence of quantum circuit execution time on parameters is non-trivial, especially when simulated on classical computing resources, making it difficult to allocate computing resources optimally during quantum circuit hyperparameter selection.

Innovation Solution

A method for predicting the training time of quantum circuit-based machine learning models using a predictor machine learning model that takes into account quantum circuit parameters such as qubit number, variational qubit gates, and entangling gates, allowing for efficient resource allocation and optimization of quantum circuit architectures.

Engineering Contradictions & Design Principles

VSEngineering Contradiction Analysis

1Adaptability or versatility

If quantum circuit architecture parameters are systematically varied to determine efficient quantum processing resources, then the ability to find optimal quantum circuit architectures is improved, but the training time and computational cost increase significantly

Engineering Contradiction:

Improvequantum circuit architecture optimizationVSAvoidtraining time

Core Design Contradiction:

Adaptability or versatilityVSLoss of time

Solution Approach 1:

The patent applies preliminary action by training a predictor machine learning model in advance to estimate training times for different quantum circuit architectures. This predictor is trained beforehand using a dataset of quantum circuit parameters and their corresponding actual training times, enabling fast predictions without executing actual training experiments for each architecture candidate.

Inventive Principle:

Principle #10Preliminary action

Solution Approach 2:

The patent uses copying by creating a surrogate model (predictor machine learning model) that replicates the behavior of actual quantum circuit training. The predictor copies the relationship between quantum circuit parameters and training time based on historical training data, allowing architects to evaluate multiple configurations quickly without running full training experiments.

Inventive Principle:

Principle #26Copying

2Power

If the number of quantum gates and entangling gates is increased to improve quantum advantage, then the computational power increases, but the execution time and resource requirements increase non-trivially

Engineering Contradiction:

Improvequantum computational powerVSAvoidquantum circuit execution time

Core Design Contradiction:

PowerVSDuration of action of moving object

Solution Approach 1:

The patent applies parameter changes by using the predictor machine learning model to evaluate how changes in quantum circuit parameters (number of gates, entangling gates, qubit connections) affect training time. This allows researchers to adjust parameters to find the optimal balance between quantum computational power and execution time before actually implementing the circuit.

Inventive Principle:

Principle #35Parameter changes

Solution Approach 2:

The patent performs preliminary evaluation of quantum circuit configurations by predicting their training times before actual implementation. This preliminary action uses the trained predictor model to assess the impact of increasing quantum gates and entangling gates on execution time, enabling informed decisions about circuit design.

Inventive Principle:

Principle #10Preliminary action

3Ease of operation

If quantum circuit simulations are performed on classical computing resources to evaluate architectures, then architecture selection is enabled, but the computational cost and time consumption become prohibitive

Engineering Contradiction:

Improvequantum circuit architecture evaluationVSAvoidcomputational efficiency

Core Design Contradiction:

Ease of operationVSProductivity

Solution Approach 1:

The patent uses copying by creating a classical machine learning predictor that replicates the functionality of quantum circuit training evaluation. Instead of performing actual quantum circuit simulations on classical resources, the predictor copies the essential relationship between circuit parameters and training time, providing a computationally efficient surrogate for architecture evaluation.

Inventive Principle:

Principle #26Copying

Solution Approach 2:

The patent applies mechanics substitution by replacing the mechanical process of actual quantum circuit simulation and training with a machine learning-based prediction system. The predictor model substitutes the computationally intensive quantum circuit execution with a fast classical inference process that estimates training times without actual quantum simulation.

Inventive Principle:

Principle #28Mechanics substitution (Replace mechanical system)

Data Source

PatentEP4703981A1Method for predicting a training time of a quantum circuit based machine learning model

Publication Date: 2026.03.04 TERRA QUANTUM AG

Figure 1
Figure 2
Figure 3

AI summary

A computer-implemented method for estimating a predicted training time for training a quantum-circuit based machine learning model comprising a variational quantum circuit, said method comprising the steps of receiving a quantum circuit architecture of the variational quantum circuit to be assessed for obtaining quantum circuit parameters of the variational quantum circuit including a qubit number, a number of variational qubit gates, and a number of entangling gates; providing the quantum circuit parameters to a predictor machine learning model trained to predict an execution time of the quantum circuit to generate a predicted execution time for the quantum circuit parameters; and determining based on the predicted execution time and a set of training hyperparameters, a training time for training the quantum circuit-based machine learning model.