Machine Learning for Materials Science

The field of materials science is witnessing a significant shift towards the adoption of machine learning techniques to accelerate the discovery, design, and optimization of novel materials. Researchers are developing innovative methods to learn from limited and complex data, enabling the prediction of material properties and behaviors with unprecedented accuracy. Notably, advances in foundation models, meta-learning, and semi-supervised learning are allowing scientists to overcome traditional limitations in data availability and quality. Some noteworthy papers in this regard include:

  • PolyMicros, which introduces a novel machine learning approach for learning from hyper-sparse data and demonstrates its utility in constructing a foundation model for polycrystalline materials.
  • Robust Molecular Property Prediction via Densifying Scarce Labeled Data, which proposes a meta-learning-based approach to improve the generalization of molecular prediction models to out-of-distribution compounds.
  • Knowledge Distillation Framework for Accelerating High-Accuracy Neural Network-Based Molecular Dynamics Simulations, which presents a novel knowledge distillation framework for training neural network potentials, achieving comparable or superior accuracy in reproducing physical properties while reducing the number of expensive calculations.

Sources

PolyMicros: Bootstrapping a Foundation Model for Polycrystalline Material Structure

Robust Molecular Property Prediction via Densifying Scarce Labeled Data

Spectra-to-Structure and Structure-to-Spectra Inference Across the Periodic Table

Accurate and Uncertainty-Aware Multi-Task Prediction of HEA Properties Using Prior-Guided Deep Gaussian Processes

When and How Unlabeled Data Provably Improve In-Context Learning

Knowledge Distillation Framework for Accelerating High-Accuracy Neural Network-Based Molecular Dynamics Simulations

Built with on top of