opacus

Dopamine: Differentially Private Federated Learning on Medical Data (AAAI - PPAI)

Easy-to-use utilities to build privacy-preserving AI.

Securing Collaborative Medical AI by Using Differential Privacy

Code for the paper "PUFFLE: Balancing Privacy, Utility, and Fairness in Federated Learning" by L. Corbucci, M. A. Heikkilä, D.S. Noguero, A. Monreale, N. Kourtellis.

Differential Privacy in Machine Learning under Membership Inference Attacks

Hands-on part of the Federated Learning and Privacy-Preserving ML tutorial given at VISUM 2022

Building an AI model for chest X-ray under patient privacy guarantees

A differentially private spiking neural network with temporal enhanced pooling

DP-SGD accuracy vs. privacy budget (ε) experiments on MNIST using Opacus — quantifies the accuracy/privacy/compute trade-off with RDP accounting.

A Comparative Study of Gradient Clipping Techniques in Differentially Private Stochastic Gradient Descent (DP-SGD)

Imagenet-scale DP-SGD training using Pytorch Opacus

Companion repository for the paper Threat-Driven Frameworks for Privacy-Preserving Machine Learning: A Practitioner’s Guide (2017–2025). Contains benchmarking tables, framework metrics, figure sources, and reference summaries for privacy-preserving ML techniques.

Audio classification research in PyTorch: Transformer models, OpenSMILE/MEL features, VIB and encoder-only DP experiments. Synthetic CPU checks and speaker-aware cross-validation. 隐私感知音频学习

🔐 IEEE SAFL framework re-implementation: Federated recommendation with DP-SGD (ε=2.58), FedProx & Sentence-BERT — 54.6% MAE reduction

SecureMed-LLM: A privacy-preserving framework for clinical report generation from chest X-rays, integrating Med-Guard anonymization, DP-SGD (ε=3.0), adversarial training, IDS-LLM validation, and ECIES/Curve25519 encryption. PeerJ Computer Science 2025.

A hands-on educational walkthrough of training a CelebA (Eyeglasses) image classifier with Differentially Private SGD using PyTorch and Opacus. The focus of this repo is on clarity and reproducibility through balanced subsets, deterministic preprocessing, and side-by-side baseline vs. DP training, while acknowledging real trade-offs.

Cross-silo federated learning on FEMNIST with membership inference attacks, Record-Level DP-SGD, Subject-Level DP, and privacy–utility analysis.

Privacy-preserving credit risk scoring API: DP-SGD + API hardening defending against membership inference and model extraction (307K records, <1% AUC drop at ε=2.0)

Differentially private synthetic tabular data. T5-small with LoRA under DP-SGD, served through FastAPI and Next.js on Cloud Run.

In this project we add differential privacy into an openset recognizer.to implement DP we use opacus library.