Towards Explainable Quantum AI: Informing the Encoder Selection of Quantum Neural Networks via Visualization

Area of Science:

• Quantum Computing
• Artificial Intelligence
• Machine Learning

Background:

• Quantum Neural Networks (QNNs) integrate quantum computation with neural networks for enhanced data processing.
• Selecting appropriate encoders for mapping classical data to quantum states in QNNs is challenging due to a lack of systematic methods.
• Current approaches rely on trial-and-error, hindered by difficulties in pre-training evaluation of quantum states and feature distinguishability analysis.

Purpose of the Study:

• To address the challenges in selecting QNN encoders.
• To introduce a novel visualization tool, XQAI-Eyes, for comparing classical data features with encoded quantum states.
• To provide a method for analyzing encoder effectiveness in distinguishing data features and examining mixed quantum states.

Main Methods:

• Development and application of the XQAI-Eyes visualization tool.
• Comparison of classical data features with corresponding encoded quantum states within QNNs.
• Examination of mixed quantum states across different data classes using XQAI-Eyes.
• Evaluation of XQAI-Eyes across diverse datasets and encoder designs.

Main Results:

• XQAI-Eyes facilitates a deeper understanding of how encoders impact QNN performance by bridging classical and quantum perspectives.
• The tool aids in exploring the relationship between encoder design and QNN effectiveness.
• Domain experts utilized XQAI-Eyes to establish two key practices for quantum encoder selection: pattern preservation and feature mapping.

Conclusions:

• XQAI-Eyes offers a holistic and transparent approach to optimizing quantum encoders for QNNs.
• The visualization tool enhances the systematic selection and understanding of quantum encoders.
• The derived practices provide practical guidance for improving QNN development and performance.