Skip to content

GraphIDS

CAN bus intrusion detection with graph–Mamba models. New learning work uses message-preserving inputs: individual CAN frames enter a bus-sequence Mamba encoder and learned causal graph readout. The snapshot implementation is archived for reproduction. The active Ray/Lightning/Slurm experiment path uses continuous event history and independent recording streams on one GPU.

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.
  • Research code style — readable scientific logic, boundary validation, and behavior-focused tests.
  • API Reference — auto-generated from docstrings.

Source