Resources¶
Last updated: 14 September 2026
Course services¶
- Codabench — prediction submissions and leaderboards
- Kaggle — notebooks with a free GPU, and public datasets
- Weights & Biases — experiment tracking
Complete the required course-service accounts by Session 3 (21 September).
Notebook computing¶
Run in-class notebooks on Google Colab or on a personal laptop.
Kaggle Notebooks are another option for running notebooks with a free GPU. In the notebook editor, open Settings → Accelerator → GPU. GPU access is subject to Kaggle's availability and usage limits.
NSTC Core¶
NSTC Core is used for homework and final project training. Apply by the same Session 3 deadline — we will walk through the application in Session 2.
Apply for an NSTC Core account
Submit your account application using the same email address registered in NTHU's Academic Information System (校務資訊系統).
Current usage limits are:
- maximum time per job: 1 hour;
- maximum queued jobs: 5;
- maximum concurrent running jobs: 1; and
- weekly allocation: 20 hours per student.
Use checkpoint-and-resume for longer training, avoid large parallel hyperparameter sweeps, and change one experimental factor at a time. Pretrained checkpoints will be provided when training from scratch is outside the assignment's compute budget.
Do not rely on a queue during class
NSTC Core is for homework and final project training, not for in-class tasks.
Further reading¶
Optional.
- Hands-On Machine Learning with Scikit-Learn and PyTorch — Aurélien Géron (O'Reilly, 2025). Classical methods in scikit-learn first, then neural networks in PyTorch.
Getting access
The NTHU Library holds this book in print and as an ebook, so there is no need to buy a copy.
Acknowledgements¶
This course is a new offering of Hands-on Artificial Intelligence (AI) 物理與人工智慧（AI）實作, conceived and taught as PHYS591000 by Prof. Pai-Hsien Jennifer Hsu and Prof. Shih-Chieh Hsu. Its structure, its choice of physics problems, and much of its teaching approach follow the path they set, and this course would not exist without their work. Their 2023 course materials remain online and are worth reading alongside this course.
The slides, notebooks, and assignments here have been rewritten for this offering; any errors in them are mine alone.
Datasets, papers, and software used in the course remain the work of their respective authors and are credited where they appear.