AIST5040 PhysAI and GenAI for Natural Science
(a.k.a. AI for Science)
-
Department of Computer Science and Engineering
The Chinese University of Hong Kong
2026-2027 Term 1
Description
This course introduces PhysAI and GenAI as complementary approaches to natural scientific discovery. PhysAI refers to physics-inspired AI models that embed domain knowledge, physical laws, and invariances into learning frameworks, enabling models to achieve greater robustness, accuracy, and interpretability. In contrast, GenAI leverages generative modeling to create hypotheses, design candidates, and explore novel scientific paradigms beyond the limits of existing data. Together, PhysAI and GenAI form a powerful synergy: PhysAI accelerates and strengthens established paradigms, while GenAI opens pathways to entirely new directions of research. Through this integration, PhysAI and GenAI hold the potential to transform discovery in chemistry, materials science, and biology.
Advisory: Students are expected to have taken AIST1000 or AIST3120 or AIST4010 or AIST4030 or AIST5030 or CSCI3230 or CSCI3320.
Reference Materials
- [Paper] Yang Song, Diederik Kingma, How to Train Your Energy-Based Models. ArXiv'21.
- [Tutorial] Physics of Language Models. ICML'24.
- [Tutorial] Physics-Inspired Geometric Pretraining for Molecule Representation. AAAI'25.
- [Tutorial] Multi-modal Foundation Model for Scientific Discovery: With Applications in Chemistry, Material, and Biology. AAAI'25.
- [Book] The Principles of Diffusion Models. ArXiv'26.
- [Book] Generative AI and Stochastic Thermodynamics. 2026.
Syllabus
- Fri 10:30am - 1:30pm, Lee Shau Kee Building 514
| Date | Topics | References & Presentation Topics | |
|---|---|---|---|
| Week 1 | Sep 11th | Overview of AI for Science | |
| Week 2 | Sep 18th |
AI & Physics Foundation: GenAI #1 |
(PhysAI for) Density Estimation
Autoregressive Models
Energy-Based Model
Contrastive Learning
Score-Based Models
Variational Model
Flow-Based Models
PhysAI for PDE
PhysAI for Symmetry
PhysAI for Interpretability
|
| Week 3 | Sep 25th |
AI & Physics Foundation: GenAI #2 PhysAI |
|
| Week 4 | Oct 2nd |
AI & Physics Foundation: Transfer Learning, Multi-modal Learning, Large Language Model, Foundation Model, and World Model #1 |
Key Architecture
ViT Series
JEPA Series
Others
|
| Week 5 | Oct 9th |
AI & Physics Foundation: Transfer Learning, Multi-modal Learning, Large Language Model, Foundation Model, and World Model #2 Guest Lecture by Ziyu Li: Topics: World Model |
|
| Week 6 | Oct 16th |
PhysAI for Chemistry: Molecular Representation and Property Prediction, Energy and Force, AI MD |
Molecular Representations (String/1D)
Molecular Property Prediction (Graph/2D)
Geometric & Equivariant Molecular Learning
Neural Potentials & Molecular Dynamics
Molecular Pretraining
Molecule-Text-3D Multimodal Learning
Quantum Chemistry, Electronic Structure &
Baselines
|
| Week 7 | Oct 23rd |
GenAI for Chemistry: Molecule Generation and Optimization, Reaction Prediction, Retrosynthesis and Synthesis Planning |
Molecular Sequence Generation
Molecular Graph Generation
Molecular Conformation Generation
Joint Molecular Structure Generation
Goal-Directed Generation & Molecular Optimization
Molecule-Text Representation, Retrieval &
Generation
Reaction Prediction & Atom Mapping
Retrosynthesis & Synthesis Planning
Chemistry LLMs & Agents
|
| Week 8 | Oct 30th |
PhysAI for Biology: Protein Structure, Genome Function, Single-Cell and Gene Regulation Guest Lecture by Yanjing Li: Topics TBD |
Protein Language Models & Representation
Learning
Protein Structure & Complex Prediction
Inverse Folding
Structure Search & Geometric Pretraining
Genome Sequence-to-Function & Variant Effects
Single-Cell & Gene Regulatory Models
|
| Week 9 | Nov 6th |
GenAI for Biology: Protein Design and Engineering, Cell-State and Perturbation Modeling, Genome and RNA Design, Structure-based Drug Design Guest Lecture by Dr. Yuning You: Topics TBD |
Protein Structure Generation
Protein Sequence Generation
Protein Conformational Ensembles
Docking & Binding
Structure-based Drug Design
Multimodal Biomolecular Design
Genome Foundation Models & Generative Design
Single-Cell Perturbation & Cell-State
Generation
RNA Structure & Design
Surveys & Benchmarks
|
| Week 10 | Nov 13th |
PhysAI for Material Science: Crystallization, Phase Detection |
Crystal Representation & Property Prediction
Interatomic Potentials & Atomistic Simulation
Phase, Symmetry & Reciprocal-Space Learning
|
| Week 11 | Nov 20th |
GenAI for Material Science: Material Generation, Structure Generation |
Unconditional Crystal Generation
Property-Conditioned & Targeted Generation
Language Models & Multimodal Materials Design
Porous Materials & MOF Design
Surveys & Benchmarks
|
| Week 12 | Nov 27th | Project Presentation #1 | |
| Week 13 | Dec 4th | Project Presentation #2 Summary |