Advancing materials science through next-generation machine learning.

Area of Science:

• Materials Science
• Artificial Intelligence
• Machine Learning

Background:

• Machine learning (ML) models have shown success in computer vision and natural language processing (NLP).
• Recent advances include large-scale language and generative image models, increasing accessibility.
• Current ML models in materials science are specialized, limiting broader industrial application.

Purpose of the Study:

• To address the limitations of specialized ML models in materials science.
• To propose the development of a comprehensive and versatile ML model for materials science.
• To enable intuitive querying and innovative solutions in materials discovery.

Main Methods:

• Leveraging representation learning, generative modeling, and foundation model techniques.
• Establishing an extensive, centralized dataset by crowdsourcing and literature data extraction.
• Training a central model on a massive dataset to learn underlying physics.

Main Results:

• Current models are overly specialized, hindering industrial process integration.
• A versatile model can interpret human-readable inputs and identify search directions.
• The proposed approach facilitates both searching existing data and innovating new solutions.

Conclusions:

• Materials science requires versatile ML models capable of understanding human queries.
• A centralized, comprehensive dataset is crucial for training such models.
• The envisioned model will enhance materials discovery by integrating existing knowledge and enabling innovation.