Generalization Performance of Quantum Metric Learning Classifiers.

Area of Science:

• Quantum computing applications in machine learning.
• Development of quantum algorithms for data classification.

Background:

• Quantum computing offers potential advantages in machine learning, particularly for kernel-based methods.
• Quantum metric learning uses quantum embeddings for data separation, successfully classifying training data.
• Previous assessments lacked evaluation of classifier performance on unseen test data.

Purpose of the Study:

• To evaluate the predictive performance of quantum metric learning on unseen test datasets.
• To investigate the impact of dimensionality reduction using principal component analysis (PCA) on quantum metric learning.
• To assess the feasibility of hybrid classical-quantum computation for high-dimensional data embedding.

Main Methods:

• Applied quantum metric learning to ant/bee image datasets and breast cancer clinical data.
• Utilized principal component analysis (PCA) for feature data dimensionality reduction prior to quantum embedding.
• Compared the original quantum metric learning approach with PCA-enhanced variants.
• Constrained the number of model parameters relative to training samples.

Main Results:

• Quantum metric learning accurately classified test data for both image and clinical datasets.
• PCA-based dimensionality reduction, when limited, improved the performance of quantum metric learning.
• Performance was successful when the number of model parameters was significantly less than the number of training samples.

Conclusions:

• Quantum metric learning, especially with PCA-enhanced dimensionality reduction, demonstrates accurate predictive capabilities on test data.
• Hybrid classical-quantum approaches are viable for embedding high-dimensional data into quantum systems.
• The number of model parameters is a critical factor for successful generalization in quantum machine learning.