About AuditML¶
Project¶
AuditML is a final-year project (FYP) developed at NUML Faisalabad. It provides an end-to-end privacy auditing toolkit for PyTorch models, covering:
- Four privacy attacks (Threshold MIA, Shadow MIA, Model Inversion, Attribute Inference)
- Differential Privacy training via Opacus
- Automated report generation with metrics and visualisations
- Self-contained HTML reports with inline charts (auto-opens in browser)
- An optional Rust acceleration module (~11× faster threshold scanning)
Author¶
Eeman Asghar
NUML Faisalabad, 2024–2025
References¶
| Paper | Relevance |
|---|---|
| Yeom et al., Privacy Risk in Machine Learning, IEEE CSF 2018 | Threshold MIA |
| Shokri et al., Membership Inference Attacks Against Machine Learning Models, IEEE S&P 2017 | Shadow MIA |
| Fredrikson et al., Model Inversion Attacks, CCS 2015 | Model Inversion |
| Abadi et al., Deep Learning with Differential Privacy, CCS 2016 | DP-SGD |
| Carlini et al., Membership Inference Attacks from First Principles, IEEE S&P 2022 | LiRA and TPR@low FPR |
Tech stack¶
- PyTorch — model training and inference
- Opacus — Differential Privacy
- scikit-learn — metrics
- MkDocs Material — documentation
- Rust / PyO3 / maturin — performance extension
License¶
MIT License. See LICENSE in the repository root.