Inference Attacks Library

Overview

Welcome to the Inference Attacks library, a comprehensive toolset designed to facilitate the implementation and analysis of privacy inference attacks on datasets. This library offers a robust and extensible framework to conduct various types of privacy attacks, providing users with insights into potential vulnerabilities within their datasets.

Installation

To integrate the Inference Attacks library into your project, use the following pip command:

pip install infattacks

Example of usage

import infattacks

# Example: Create a Dataset instance from a CSV file
dataset = infattacks.Data(file_name="path/to/your/dataset.csv", sep_csv=",")

# Attack
qids = ["age", "education"]
sensitive = ["income"]

attack = infattacks.Probabilistic(dataset, qids=qids, sensitive=sensitive)
prior_reid = attack.prior_reid()
prior_ai = attack.prior_ai()
exp_reid, hist_reid = attack.post_reid(qids=qids, hist=True)
exp_ai, hist_ai = attack.post_ai(qids=qids, hist=True)

print("[Re-identification]")
print("Prior vulnerability: %.6f"%(prior_reid))
print("Posterior vulnerability: %.6f"%(exp_reid))

print("\n[Attribute-Inference]")
for att in sensitive:
    print("Prior vulnerability [%s]: %.6f"%(att, prior_ai[att]))
    print("Posterior vulnerability [%s]: %.6f"%(att, exp_ai[att]))

vis = infattacks.VisualizeRisk(attack)
vis.plot_hist(hist_reid, "Re-identification - QIDs: [age,education]")
vis.plot_hist(hist_ai["income"], "Attribute-Inference - QIDs: [age,education] - sensitive: income")

License

This library is distributed under the MIT License. Feel free to use, modify, and distribute it according to the terms outlined in the license.