Contrastive Metric Learning for Lithium Super-ionic Conductor Screening

Area of Science:

• Materials Science
• Electrochemistry
• Computer Science

Background:

• High-performance lithium-ion batteries are crucial for modern technology.
• Materials with high electrical conductivity are essential for advanced battery development.
• Predicting high conductors is challenging due to a lack of validated samples.

Purpose of the Study:

• To develop an effective metric-learning framework for screening high-conductivity materials.
• To address the challenge of limited validated conductor samples in machine learning models.
• To improve the prediction accuracy of conductive materials for battery applications.

Main Methods:

• Utilized a Siamese network to map material structures into an optimized feature space.
• Employed an instance-based method for classifying input material samples.
• Developed a metric-learning framework for direct high conductor screening.

Main Results:

• The proposed method effectively extracts knowledge from imbalanced datasets.
• Demonstrated good performance and generalization ability in conductor screening.
• Successfully mapped material structures to a discriminative feature space.

Conclusions:

• The metric-learning framework offers a viable solution for high conductor screening.
• The approach shows promise for accelerating the discovery of advanced battery materials.
• Effective handling of imbalanced data is a key strength of the proposed method.