MPpredictor: An Artificial Intelligence-Driven Web Tool for Composition-Based Material Property Prediction.

Area of Science:

• Materials Science
• Computational Materials Science
• Artificial Intelligence in Materials

Background:

• AI, ML, and DL are increasingly vital in materials science for data-driven insights.
• These techniques accelerate the discovery and design of novel materials.
• Extracting and utilizing data efficiently is key to advancing materials research.

Purpose of the Study:

• To deploy predictive models for multiple material properties based on material composition.
• To assist in the acceleration of materials discovery and design.
• To provide an accessible online tool for predicting material properties.

Main Methods:

• Utilized deep learning models built with a cross-property deep transfer learning technique.
• Leveraged source models trained on large datasets to build target models on smaller datasets.
• Developed an online software tool for material property prediction.

Main Results:

• The developed models predict up to 41 different material property values.
• The online tool preprocesses material compositions to generate relevant attributes.
• Predictive models successfully forecast material properties from input compositions.

Conclusions:

• Deep transfer learning is effective for predicting material properties from composition.
• The online tool facilitates rapid prediction of numerous material properties.
• This approach aids in accelerating materials discovery and design for various applications.