ege-erdogan/unsplit

Supplementary code for the paper "UnSplit: Data-Oblivious Model Inversion, Model Stealing, and Label Inference Attacks Against Split Learning".

27
/ 100
Experimental

This project helps machine learning engineers and researchers evaluate the privacy risks in split learning systems. It takes a trained split learning model and demonstrates how an untrusted server can reconstruct sensitive input data, steal the client's model, and infer training labels. The output highlights vulnerabilities, showing that split learning may offer a false sense of security regarding data privacy.

No commits in the last 6 months.

Use this if you are designing or deploying a split learning system and need to understand its potential privacy weaknesses against an honest-but-curious server.

Not ideal if you are looking for methods to enhance the privacy of your split learning setup, as this project focuses on demonstrating vulnerabilities, not providing defenses.

split-learning privacy-assessment model-security federated-learning data-confidentiality
No License Stale 6m No Package No Dependents
Maintenance 0 / 25
Adoption 5 / 25
Maturity 8 / 25
Community 14 / 25

How are scores calculated?

Stars

14

Forks

3

Language

Python

License

Last pushed

Nov 10, 2022

Commits (30d)

0

Get this data via API

curl "https://pt-edge.onrender.com/api/v1/quality/ml-frameworks/ege-erdogan/unsplit"

Open to everyone — 100 requests/day, no key needed. Get a free key for 1,000/day.