AI and Physics

Machine Learning for Materials Science

My recent work in "Explainable AI for Curie Temperature Prediction in Magnetic Materials" demonstrates the power of combining domain expertise in physics with modern machine learning techniques. By augmenting the NEMAD database with composition-based and domain-aware descriptors, we developed models that not only achieve high predictive accuracy (R² = 0.85) but also provide interpretability through SHAP analysis.

Key Contributions

Predictive Modeling: Evaluated multiple machine learning models, with Extra Trees Regressor delivering the best performance
Explainability: Used SHAP analysis to identify key physicochemical drivers such as average atomic number and magnetic moment
Clustering Insights: Employed k-means clustering to understand performance across chemically distinct material groups
Practical Impact: Provides tools for accelerating the discovery of magnetic materials with desired Curie temperatures

This work exemplifies how AI can augment traditional physics approaches, enabling rapid screening of material properties while maintaining physical interpretability—crucial for scientific discovery and industrial applications.

Read the Full Paper