Overview
The Norwegian AI Cloud promised that shared infrastructure could lower the barrier for researchers to apply machine learning in their own domains. Work Package 7 tested that promise by building seven demonstrator projects, each solving a real research problem and each packaged as a self-contained pipeline that runs on NAIC Orchestrator VMs.
The seven use cases cover a wide range of AI techniques and scientific disciplines. Some, like the PEM electrolyzer optimizer (UC2) and pseudo-Hamiltonian networks (UC3), explore how embedding physics into neural networks can produce models that are smaller and generalize better than pure data-driven approaches. Others, like the climate teleconnection analysis (UC1) and AIS vessel classification (UC5), tackle large-scale data-driven discovery, finding patterns that would be impractical to identify manually. The rest address foundational computational challenges: multi-modal optimization (UC6), latent PDE representations (UC7), and medical image registration (UC4).
They all share a common workflow: each demonstrator ships a Jupyter notebook that a researcher can provision on an NAIC Orchestrator VM and run end-to-end, from data loading through training to results, without managing infrastructure.
Use Case Demonstrators
Use Case |
Title |
Repository |
Tutorial |
DOI |
Status |
|---|---|---|---|---|---|
UC1 |
Climate Indices Teleconnection |
Completed |
|||
UC2 |
PEM Electrolyzer PINN Optimizer |
Completed |
|||
UC3 |
Pseudo-Hamiltonian Neural Networks |
Completed |
|||
UC4 |
3D Medical Image Registration |
Completed |
|||
UC5 |
Graph-Based AIS Classification |
Completed |
|||
UC6 |
Multi-Modal Optimization |
Completed |
|||
UC7 |
Latent Representation of PDE Solutions |
Completed |
Summary Deliverables
ID |
Deliverable |
Repository |
DOI |
|---|---|---|---|
D7.10 |
Summary of completed demonstrators |
Getting Started
Each demonstrator repository includes everything needed to go from git clone to results:
Provision a VM at orchestrator.naic.no
Clone the repository and run
setup.shto configure the environmentOpen the orchestrator notebook in JupyterLab and run end-to-end
All demonstrators include Sphinx-based tutorials published on GitHub Pages, and six (all except UC4) include AGENT.md files that let AI coding assistants set up and run the project autonomously.
Technology Stack
Technology |
Use Cases |
Role |
|---|---|---|
Python |
All |
Primary implementation language |
PyTorch |
UC2, UC3, UC5 |
Deep learning framework |
TensorFlow |
UC7 |
Deep learning framework |
scikit-learn / XGBoost |
UC1 |
Classical ML and gradient boosting |
DGL |
UC5 |
Graph neural network library |
ANTsPy / HD-BET |
UC4 |
Medical image registration and brain extraction |
CMA-ES / DEAP |
UC6 |
Evolutionary optimization |
Infrastructure
Infrastructure |
Use Cases |
Purpose |
|---|---|---|
NAIC Orchestrator VMs |
All |
GPU-enabled cloud VMs for interactive development and training |
Jupyter Notebooks |
All |
Interactive demonstrator interfaces |
Sphinx Tutorials (GitHub Pages) |
All |
Multi-chapter tutorial documentation |
AI Agent Files |
UC1, UC2, UC3, UC5, UC6, UC7 |
AI coding assistant integration (AGENT.md/AGENT.yaml) |