Code for: Detection-Aware Content-Level Differential Privacy for Lattice-Trapdoor Traceable Federated Prototype Learning

Abstract

This repository contains the runnable implementation and experiment scripts accompanying the paper "Detection-Aware Content-Level Differential Privacy for Lattice-Trapdoor Traceable Federated Prototype Learning". It implements a lattice-trapdoor identity-tracing mechanism for federated prototype learning (FedLHT) combined with a content-level differential-privacy (DP) layer and a detection-aware release ordering that prevents the anomaly detector from collapsing under DP noise.Included: the lattice-trapdoor signature scheme, DP clipping/noise and RDP accountant, the federated prototype-learning simulator, privacy-attack implementations (prototype inversion, membership inference, property inference), and scripts to reproduce the main utility-under-attack sweep, privacy-attack evaluation, overhead benchmark, cryptographic correctness self-test, and robustness checks reported in the paper.See README.md for setup instructions and the exact command to run each experiment.