Exploring quantum active learning for materials design and discovery

Area of Science:

• Materials Science
• Quantum Computing
• Artificial Intelligence

Background:

• Classical active learning (AL) has demonstrated efficiency in data usage for materials discovery.
• Quantum machine learning (QML) offers potential for developing advanced regression models.
• The synergy between AI and quantum computing is a rapidly advancing field.

Purpose of the Study:

• To explore the application of quantum algorithms within the active learning (AL) framework for materials discovery.
• To investigate the performance of quantum active learning (QAL) using quantum support vector regressors (QSVR) and quantum Gaussian process regressors (QGPR).
• To compare QAL performance against classical AL in materials property prediction and structure optimization.

Main Methods:

• Implementation of QAL using MLChem4D and QMLMaterial codes.
• Utilization of QSVR and QGPR with diverse quantum kernels and feature maps.
• Application to datasets including perovskite properties (piezoelectric coefficient, band gap, energy storage) and nanoparticle structure optimization via density functional theory.

Main Results:

• QAL improved search efficiency in most tested cases.
• Performance variations in QAL were observed, potentially linked to data "roughness."
• The study validated the potential of QAL for discovering optimal solutions within chemical spaces.

Conclusions:

• QAL represents a promising advancement for data-efficient materials discovery.
• The integration of quantum chemistry with QML, termed the "QQ method," facilitates new inferences and discoveries.
• QAL holds significant potential for applications in materials science, chemistry, and beyond.