A repetitive amplitude encoding method for enhancing the mapping ability of quantum neural networks

Area of Science:

• Quantum Computing
• Machine Learning
• Artificial Intelligence

Background:

• Quantum neural networks (QNNs) are a rapidly developing area in quantum machine learning.
• Current QNNs face limitations in mapping capability due to the linear nature of quantum gates used for feature mapping.
• Enhancing the mapping capability of QNNs is crucial for their broader applicability.

Purpose of the Study:

• To propose and evaluate a novel repetitive amplitude encoding method for QNNs.
• To improve the data mapping capability of QNNs beyond existing linear transformations.
• To demonstrate the superior performance of the proposed method compared to traditional encoding techniques.

Main Methods:

• A repetitive amplitude encoding method is introduced, which encodes probability amplitudes of multiple qubit blocks by reusing classical data.
• The proposed method was tested using the MNIST dataset to compare its performance against existing encoding methods.
• The effectiveness was further validated on reservoir lithology identification, IRIS, and WINe classification datasets.

Main Results:

• Repetitive amplitude encoding demonstrated superior performance over other methods when the number of classes was fixed.
• As the number of classes increased with a fixed number of hidden layers, the performance advantage of repetitive amplitude encoding became more pronounced.
• The proposed QNN method showed adaptability and superior classification performance compared to classical neural networks across diverse datasets.

Conclusions:

• The repetitive amplitude encoding method significantly enhances the mapping capability of QNNs.
• This novel approach offers a promising solution for improving QNN performance in various classification tasks.
• The method shows potential for practical applications in fields like oil and gas exploration and general machine learning.