Methods
Machine learning methods commonly used in scientific applications.
Overview
Scientific ML draws from various methodologies:
- Graph Neural Networks — For molecular and materials data
- Transformers & LLMs — Language models for science
- Generative Models — Designing new molecules and materials
- Active Learning — Efficient experimental design
- Uncertainty Quantification — Knowing what we don’t know
Quick Start
| Method | Domain Examples |
|---|---|
| Graph Neural Networks | Molecular property prediction, materials |
| Transformers | Reaction prediction, property prediction |
| Generative Models | Molecule design, materials discovery |
| Active Learning | Self-driving labs, experimental optimization |
Resources by Method
Graph Neural Networks
| Resource | Focus |
|---|---|
| DeepChem Tutorials | GNNs for molecules |
| PyTorch Geometric | General GNN framework |
Transformers & LLMs
| Resource | Focus |
|---|---|
| Transformers for Chemistry | LLMs in chemistry/materials |
| awesome-scientific-language-models | Scientific LLMs |
Generative Models
| Resource | Focus |
|---|---|
| AI4Chemistry Course | Generative models for chemistry |
Conferences
| Conference | Focus | Timing |
|---|---|---|
| NeurIPS | AI4Science workshop | December |
| ICML | ML4Science workshops | July |
| ICLR | Science-focused papers | May |