Quantum support vector machine based on regularized Newton method.

Area of Science:

• Quantum Computing
• Machine Learning
• Computational Science

Background:

• Classical support vector machines (SVMs) are widely used but can be computationally intensive.
• Existing quantum SVMs, like those using the HHL algorithm, face limitations with certain input matrix structures.
• Iterative methods, such as Newton's method, are effective for training classical SVMs.

Purpose of the Study:

• To develop a quantum support vector machine (QSVM) that overcomes the input matrix constraints of prior quantum algorithms.
• To achieve exponential speed-up compared to classical SVM training and classification.
• To enhance accuracy, robustness, and computational complexity in quantum machine learning.

Main Methods:

• Introduction of a novel regularized quantum Newton algorithm to address input matrix limitations.
• Training the regularized Newton quantum support vector machine (RN-QSVM) using the developed quantum Newton algorithm.
• Classification of query samples by constructing quantum states within the RN-QSVM framework.

Main Results:

• The proposed RN-QSVM demonstrates an exponential speed-up over classical SVM algorithms.
• RN-QSVM shows superior performance in accuracy and robustness compared to QSLS-SVM, LS-QSVM, and classical methods.
• The regularized quantum Newton algorithm effectively removes constraints related to the input matrix structure.

Conclusions:

• The RN-QSVM presents a significant advancement in quantum machine learning, offering substantial speed and performance benefits.
• This quantum approach provides a more versatile and efficient alternative for SVM tasks, especially with challenging data structures.
• The study highlights the potential of integrating iterative classical optimization techniques with quantum computation for enhanced machine learning.