PrivJail Benchmarks

This repository contains benchmarks for evaluating the utility and performance of PrivJail.

Benchmarks

• Decision Tree: DP decision tree construction on the Adult dataset. Based on the paper Data Mining with Differential Privacy.
• Drive Heatmap: DP spatial heatmap generation from taxi trajectories (T-Drive).
• AIM: DP synthetic data generation using the Adaptive and Iterative Mechanism on the Adult dataset. Adapted from ryan112358/mbi.
• DP-SGD: DP training of a 2-layer MLP with DP-SGD on MNIST. Based on the paper Deep Learning with Differential Privacy.

How to run

Install uv: https://docs.astral.sh/uv/getting-started/installation/

Prepare:

./prepare_dataset.bash

Run:

./run_all.bash

Then, the results will be recorded at the ./results dir.

Postprocess:

./analyze_all.bash

License

This project is licensed under the Apache License 2.0 - see the LICENSE file for details.

Copyright 2026 TOYOTA MOTOR CORPORATION.