GraphIDS

CAN bus intrusion detection via a 3-stage knowledge distillation chain: VGAE (unsupervised reconstruction) → GAT (supervised classification) → fusion. Large models compress into small models via KD auxiliaries for edge deployment.

Where to start

• Module responsibilities — one-page map of what every layer owns from experiment YAML through runtime execution.
• Config system — how an experiment YAML becomes a cache build, training run, or analysis job.
• Data architecture — raw rows, explicit representations, materialized views, and discovery/hypotheses.
• Decisions — the ADR log. Permanent verdicts on the tools and patterns that got adopted or rejected.
• API Reference — auto-generated from docstrings.

Source

• Repo: https://github.com/frenken-lab/graphids
• Runtime docs (this site): built from docs/.