Unit 3 — Machine Learning for Nanomaterials

Overview

Classical and deep learning algorithms applied to prediction of nanomaterial properties.

Topics

• Classical algorithms: Random Forest, SVM, Gradient Boosting, regression
• Deep neural networks in PyTorch (symbolic backpropagation)
• Graph Convolutional Networks (GCN) for crystal structures
• Transfer Learning and Knowledge Distillation
• Reinforcement Learning (DQN) for nano-optimization
• Molecular descriptors and feature engineering

Key Technologies

• PyTorch — Deep learning framework
• scikit-learn — Classical ML algorithms
• RDKit — Molecular descriptors
• Plotly — Interactive visualizations

Learning Outcomes

1. Train models to predict band gap, formation energy, and surface area
2. Implement GCN for crystal structure property prediction
3. Apply Transfer Learning from pre-trained models
4. Use DQN for nanoparticle geometry optimization

Notebooks

Open in GitHub