PolyCL: contrastive learning for polymer representation learning via explicit and implicit augmentations.

Area of Science:

• Materials Science
• Computational Chemistry
• Machine Learning

Background:

• Polymers are vital in numerous applications due to their versatile properties.
• Accurate polymer representations are essential for computational design and screening using machine learning.
• Representation quality directly impacts the success of computational polymer methods.

Purpose of the Study:

• To develop a self-supervised contrastive learning paradigm, PolyCL, for learning high-quality polymer representations.
• To improve polymer representation learning without relying on labeled data.
• To identify optimal augmentation strategies for contrastive learning in polymers.

Main Methods:

• Implemented a self-supervised contrastive learning framework named PolyCL.
• Utilized a combination of explicit and implicit data augmentation strategies.
• Evaluated PolyCL's performance as a feature extractor in transfer learning tasks.

Main Results:

• PolyCL achieved superior or competitive performance on transfer learning tasks.
• The model effectively learned robust polymer representations without labeled data.
• Extensive analysis identified optimal augmentation combinations for PolyCL.

Conclusions:

• PolyCL offers an effective approach for learning high-quality polymer representations.
• The method enhances machine learning applications in computational polymer science.
• Self-supervised contrastive learning with optimized augmentation is powerful for polymer informatics.