ACUTE
APT cyberattack classifier using GRU/LSTM models trained on MITRE ATT&CK TTP sequences. Automated pipeline — register a dataset, training distributes across the cluster automatically.
Manage Datasets
Data
Register training datasets, track lineage and metadata, browse the dataset library, and see which experiments have run against each file.
- ✓Browse registered datasets
- ✓Noise level and subsequence metadata
- ✓Row counts and dataset hashes
- ✓Filter experiments by dataset
Distributed Experiments
Training
GRU and LSTM models trained automatically across the three-machine cluster. Drop a dataset in acute/data/ and the watcher handles the rest.
- ✓Automatic job dispatch via RabbitMQ
- ✓iMac + Mac Mini + MacBook workers
- ✓Solo-pool PyTorch — MPS safe
- ✓Results written to Supabase in real time
Analysis & Comparison
Results
Drill into experiment metrics — validation accuracy, EDS, per-length accuracy L1–L5. Side-by-side comparison across any set of runs.
- ✓Val accuracy and EDS metrics
- ✓Per-length accuracy breakdown
- ✓Multi-experiment comparison view
- ✓Per-class performance metrics
Doctoral research project — University of Rennes · ~90% APT identification accuracy after 3–4 TTPs observed at 90% noise