Autonomous generation of single photon emitting materials

Area of Science:

• Materials Science
• Quantum Computing
• Machine Learning

Background:

• Machine learning model training heavily relies on data quantity and quality.
• Atomistic datasets in materials science are scarce, leading to training biases.
• Existing experimental data for materials discovery is often biased towards favorable outcomes.

Purpose of the Study:

• To develop a SMILES-based variational autoencoder model for generating synthetic quantum material datasets.
• To focus on creating datasets for quantum sensing applications, specifically two-level quantum molecules with dipole blockade.
• To address data continuity issues and biases in materials discovery datasets.

Main Methods:

• Utilized a variational autoencoder (VAE) model trained on SMILES strings.
• Developed an improved sampling algorithm incorporating newly generated data.
• Applied the technique to a small initial dataset of 8000 materials.

Main Results:

• Generated over 1,000,000 candidate quantum materials.
• Created a more normally distributed synthetic dataset through an enhanced sampling algorithm.
• Identified several iodine-containing molecules as promising single-photon emitters.

Conclusions:

• The VAE-based approach effectively generates large-scale synthetic datasets for quantum materials.
• The method overcomes data scarcity and bias issues in materials discovery.
• The generated dataset highlights potential candidates for quantum sensing applications.