Home

Awesome

Analyzing the Leaky Cauldron

The goal of this project is to evaluate the privacy leakage of differential private machine learning algorithms.

The code has been adapted from the code base of membership inference attack work by Shokri et al.

Below we describe the setup and installation instructions. To run the experiments for the following projects, refer to their respective README files (hyperlinked):

Software Requirements

Installation Instructions

Assuming the system has Ubuntu 18.04 OS. The easiest way to get Python 3.8 is to install Anaconda 3 followed by installing the dependencies via pip. The following bash code installs the dependencies (including scikit_learn, tensorflow>=2.4.0 and tf-privacy) in a virtual environment:

$ python3 -m venv env
$ source env/bin/activate
$ python3 -m pip install --upgrade pip
$ python3 -m pip install --no-cache-dir -r requirements.txt

Furthermore, to use cuda-compatible nvidia gpus, the following script should be executed (copied from Tensorflow website) to install cuda-toolkit-11 and cudnn-8 as required by tensorflow-gpu:

# Add NVIDIA package repositories
$ wget https://developer.download.nvidia.com/compute/cuda/repos/ubuntu1804/x86_64/cuda-ubuntu1804.pin
$ sudo mv cuda-ubuntu1804.pin /etc/apt/preferences.d/cuda-repository-pin-600
$ sudo apt-key adv --fetch-keys https://developer.download.nvidia.com/compute/cuda/repos/ubuntu1804/x86_64/7fa2af80.pub
$ sudo add-apt-repository "deb https://developer.download.nvidia.com/compute/cuda/repos/ubuntu1804/x86_64/ /"
$ sudo apt-get update

$ wget http://developer.download.nvidia.com/compute/machine-learning/repos/ubuntu1804/x86_64/nvidia-machine-learning-repo-ubuntu1804_1.0.0-1_amd64.deb

$ sudo apt install ./nvidia-machine-learning-repo-ubuntu1804_1.0.0-1_amd64.deb
$ sudo apt-get update

$ wget https://developer.download.nvidia.com/compute/machine-learning/repos/ubuntu1804/x86_64/libnvinfer7_7.1.3-1+cuda11.0_amd64.deb
$ sudo apt install ./libnvinfer7_7.1.3-1+cuda11.0_amd64.deb
$ sudo apt-get update

# Install development and runtime libraries (~4GB)
$ sudo apt-get install --no-install-recommends \
    cuda-11-0 \
    libcudnn8=8.0.4.30-1+cuda11.0  \
    libcudnn8-dev=8.0.4.30-1+cuda11.0

# Reboot. Check that GPUs are visible using the command: nvidia-smi

# Install TensorRT. Requires that libcudnn8 is installed above.
$ sudo apt-get install -y --no-install-recommends libnvinfer7=7.1.3-1+cuda11.0 \
    libnvinfer-dev=7.1.3-1+cuda11.0 \
    libnvinfer-plugin7=7.1.3-1+cuda11.0

Obtaining the Data Sets

Data sets can be obtained using the preprocess_dataset.py script provided in the extra/ folder. The script requires raw files for the respective data sets which can be found online using the following links:

$ python3 crawl_census_data.py

Once the source files for the respective data set are obtained, preprocess_dataset.py script would be able to generate the processed data set files, which are in the form of two pickle files: $DATASET_feature.p and $DATASET_labels.p (where $DATASET is a placeholder for the data set file name). For Purchase-100X, $DATASET = purchase_100. For Texas-100X, $DATASET = texas_100_v2. For Census19, $DATASET = census.

$ python3 preprocess_dataset.py $DATASET --preprocess=1

Alternatively, Census19 data set (as is used in the attribute inference paper) can also be found in the dataset/ folder in zip format.

For pre-processing other data sets, bound the L2 norm of each record to 1 and pickle the features and labels separately into $DATASET_feature.p and $DATASET_labels.p files in the dataset/ folder (where $DATASET is a placeholder for the data set file name, e.g. for Purchase-100 data set, $DATASET will be purchase_100).