Project: Generalization Capability and Information Bottleneck Optimization for Quantum Kernel Methods

Exploring the intersection of Quantum Information Bottleneck (QIB) theory and Machine Learning generalization bounds.

• Kernel Construction: Exploring embedding-based methods (e.g., Havlíček map) and measurement-based approaches to construct robust quantum kernels.
• Information Theory: Utilizing Quantum Mutual Information (QMI) to quantify redundant information and analyze its impact on generalization error bounds.
• Optimization: Incorporating the QIB principle to address the “excessive expressiveness” issue.
• Outcome: Created projected kernels with structural bottlenecks, employing active dimensionality reduction to enhance performance on unseen data.