cassandra-anon/cassandra-bce-tram2

CASSANDRA — BCE configuration on TRAM2

Fine-tuned CTI-BERT models for extracting MITRE ATT&CK techniques from cyber threat intelligence (CTI) reports. This repository contains the BCE configuration of the CASSANDRA recipe trained on TRAM2 (50 ATT&CK sub-techniques), comprising 3 ensemble members trained with seeds {42, 123, 456}.

Anonymous artifact for ACM CCS 2026 review. Final author identification will be added after review.

Headline result

On the TRAM2 test set (30 scored documents):

• 3-seed ensemble per-document F1 (τ=0.5): 73.87%
• Exceeds Llama 3.1 8B (72.50%, Buchel et al. 2025) at 73× fewer parameters.

The per-seed table below shows the live artifact's individual seed F1s and ensemble F1; small variance from the headline (≤0.3 F1) reflects inference-time floating-point ordering on different hardware. Full per-seed and ensemble metrics are in results.json.

Architecture

LabelAttentionClassifier: a 110M-parameter CTI-BERT encoder followed by a per-label attention head.

• Encoder: ibm-research/CTI-BERT (110M params, 768 hidden)
• Head: 50 learned 768-dim label queries that attend over the encoder's last_hidden_state, followed by a shared 1-output linear layer applied per-label
• Loss: BCE with pos_weight=5.0
• Regularization / training tricks: layer-wise learning rate decay (α=0.85), exponential moving average (β=0.999), multi-seed probability averaging at inference

The architecture is custom (not derived from transformers.PreTrainedModel), so loading requires the modeling.py file shipped with this repo.

Training data

• TRAM2 (Threat Report ATT&CK Mapping v2): 151 reports, 19,178 sentences, 50 ATT&CK sub-techniques. Mean of ~82 positive examples per class.
• Splits: report-level train/test split from Buchel et al. (2025) "SoK: A Survey of Approaches for ATT&CK Classifier Construction" (120 train reports, 31 test reports — one test report excluded from per-document F1 due to empty in-vocabulary ground truth).
• Validation: 80:20 sentence-level random split within the training reports for early stopping and threshold selection.

Intended use

Map free-text CTI sentences (analyst reports, incident write-ups, vendor advisories) to ATT&CK techniques. The model takes a single sentence and outputs a probability for each of 50 techniques.

Aggregation to document level (paper convention): apply per-sentence inference, take the per-class max across sentences in a document, threshold that, report the union of predicted techniques per document. F1 is computed against the document-level technique set.

Limitations:

• Trained on English-language CTI; behavior on other languages is not characterized.
• The label vocabulary is fixed at the 50 TRAM2 sub-techniques.
• Within TRAM2, the rarest techniques have ~7 positive examples; predictions for these classes are noisier than for densely-populated techniques.

How to load and run

from modeling import load_ensemble, predict_ensemble
import os, glob

seed_dirs = sorted(glob.glob(os.path.join(os.path.dirname(__file__), "seeds", "seed-*")))
seeds = load_ensemble(seed_dirs, device="cuda")

sentences = [
    "The malware uses Windows Command Shell to execute encoded scripts.",
    "After initial access, persistence was established via Registry Run Keys.",
]
results = predict_ensemble(seeds, sentences, threshold=0.5)
for sentence, techniques in results:
    print(sentence, "->", techniques)

A complete CLI example is in inference_example.py:

pip install -r requirements.txt
python inference_example.py

Per-seed members

Seed	Per-document F1 (τ=0.5)	Selected weights
42	73.78%	EMA
123	71.97%	EMA
456	75.59%	EMA
3-seed ensemble	73.87%	—

For verification without re-running the model, each seed directory contains a seed_probs.npz file with the model's per-sentence sigmoid probabilities on the test and dev splits — sufficient to recompute every F1 number in the model card.

Citation

@inproceedings{cassandra2026,
  title  = {CASSANDRA: How Many Parameters Suffice to Automate TTP Extractions from CTI Reports---Pushing Towards the Lower Bound},
  author = {Anonymous},
  booktitle = {Proceedings of the 2026 ACM SIGSAC Conference on Computer and Communications Security (CCS)},
  year   = {2026},
  note   = {Under review — anonymous submission}
}

Please also cite the TRAM2 dataset and the CTI-BERT encoder.

License

MIT — see LICENSE.