FAQ

Objectives

Find answers to common questions about WP7 demonstrators
Troubleshoot common setup and runtime issues
Understand licensing, contributions, and future work

General

Which use case should I start with?

It depends on your background:

Background	Recommended Starting Point
Climate science	UC1 — Climate Teleconnection Analysis
Energy / green hydrogen	UC2 — PEM Electrolyzer PINN Optimizer
Physics / dynamical systems	UC3 — Pseudo-Hamiltonian Neural Networks
Medical imaging	UC4 — 3D Medical Image Registration
Maritime / geospatial	UC5 — Graph-Based AIS Classification
Optimization / operations research	UC6 — Multi-Modal Optimization
Scientific computing / PDEs	UC7 — Latent PDE Representations

If you want the most polished experience, start with UC1, UC2, or UC6 — these have the most detailed multi-chapter tutorials (8–10 episodes each).

Do I need a GPU?

Not always:

UC	GPU Required?
UC1	No (CPU sufficient for all models)
UC2	Recommended (faster training)
UC3	Recommended
UC4	No (CPU sufficient)
UC5	Recommended (CUDA 11.8)
UC6	No (CPU, but multi-core parallelization helps)
UC7	Recommended

Can I run these on my laptop?

Yes, for smaller-scale experiments. The notebooks are designed to run on NAIC Orchestrator VMs for full-scale training, but most include options for reduced parameter sweeps or smaller datasets that work on a laptop.

How do I use an AI coding assistant?

Six repositories (UC1, UC2, UC3, UC5, UC6, UC7) include AGENT.md files. Tell your assistant:

“Read AGENT.md and help me run the demonstrator.”

The assistant will set up the environment and run experiments automatically.

Infrastructure

How do I get access to NAIC Orchestrator?

Register with MyAccessID using your Feide account
Login to orchestrator.naic.no
Create a VM (see Episode 02)

If your institute is not registered, contact support@naic.no.

My SSH connection is refused — what do I do?

Check these common causes:

IP changed: If you switched networks (home/office/VPN), add your new IP in the Orchestrator
Key permissions: Ensure chmod 600 your-key.pem (Linux/Mac) or restrict via icacls (Windows)
VM expired: VMs are short-lived — check if your duration has expired

How do I access Jupyter on the VM?

Create an SSH tunnel:

ssh -i /path/to/key.pem -L 8888:localhost:8888 ubuntu@<VM_IP>

Then open http://localhost:8888 in your browser.

Technical

Why do some tests skip in CI?

Tests are designed to skip gracefully when optional heavy dependencies are unavailable:

dgl (UC5) — requires CUDA and C++ graphbolt library
torch (UC2, UC3) — installed separately with CPU-only wheels in CI
requests, fastapi (UC2) — only needed for digital twin smoke tests
antspyx, hd-bet (UC4) — heavy medical imaging dependencies

This ensures CI pipelines pass quickly while still validating project structure and code logic.

What Python version is required?

All repositories target Python 3.11. CI pipelines use the python:3.11-slim Docker image. UC4 uses Conda with Python 3.11 pinned in environment.yml.

How are the repositories licensed?

Most repositories use dual licensing:

CC BY-NC 4.0 for content (tutorials, documentation, notebooks)
GPL-3.0-only for code

UC7 currently uses MIT; this will be updated for consistency.

Contributing

How can I contribute to a use case?

Fork the repository on GitHub (NAICNO organization)
Create a feature branch
Follow the existing code style and test patterns
Submit a pull request

Who do I contact for questions?

UC	Contact
UC1	Klaus Johannsen (NORCE)
UC2	Hasan Asyari Arief (NORCE)
UC3	Sølve Eidnes (SINTEF)
UC4	Saruar Alam (UiB)
UC5	Xue-Cheng Tai (NORCE)
UC6	Klaus Johannsen (NORCE)
UC7	Klaus Johannsen (NORCE)

For infrastructure issues: support@naic.no

Keypoints

Start with UC1, UC2, or UC6 for the most detailed tutorials (8–10 episodes)
GPU is recommended but not required for most use cases
AGENT.md files enable AI coding assistants to set up and run demonstrators
Tests skip gracefully when heavy dependencies are unavailable
All repos target Python 3.11 with lean CI pipelines