Implementation of "ezDPS: An Efficient and Zero-Knowledge Machine Learning Inference Pipeline"

In this documentation, we provide an instruction how to run our code as follows.

First, install all the neccessary software and libraries

1. Anaconda

   Download and install anaconda from www.anaconda.com. Set it to the environment path.

2. Python v3.6

   • Create the python virtual environment

     conda create -n [environment_name] python=3.7
     

   • Activate the python environment

     conda activate [environment_name]
     

3. scikit-learn (v. 1.0.2)

   conda install scikit-learn==1.0.2
   

4. numpy (v. 1.21.5)

   conda install numpy==1.21.5
   

5. pandas (v. 1.1.5)

   conda install pandas==1.1.5
   

6. pywavelets (v. 1.3.0)

   conda install pywavelets==1.3.0
   

7. Rust environment (v. 1.56.0-nightly)

   • install rust

     curl --proto '=https' --tlsv1.2 -sSf https://sh.rustup.rs | sh
     

   • set rust==nightly as your default toolchain. The stable toolchain will not work with some dependencies.

     rustup default nightly && rustup update
     

To Test the Accuracy with Cross-Validation (Table 4 in the paper):

1. Go to the correct directory for the dataset to be tested
   • For UCR-ECG dataset

     cd UCR
     

   • For LFW dataset

     cd LFW
     

   • For Cifar-100 dataset
     • Since github only permits 100 MB maximum file size, we had to store our test samples on Google Drive. So, please download them from (https://drive.google.com/drive/folders/1vj81b2qCxCfQ1k6tmX8ImcB7YAXkA5s6?usp=sharing)
     • Copy the downloaded files (meta, test, train) to our CIFAR folder

       cp meta test train CIFAR/cifar-100-python/
       

     • Then go to CIFAR folder

       cd CIFAR
       

2. Run the following command to obtain the result

   • For decision tree

     python train_dt.py {#class} 
     

     where {#class} is the number of classes. For UCR-ECG, replace {#class} with {8, 16, 32, or 42}. For Cifar-100, replace with {8, 16, 32, 64, 100}. For LFW, replace with {8, 16, 32, 64, 128}.

   • For DWT+PCA+DT

     python train_dwt_pca_dt.py {#class} 
     

   • For SVM only

     python train_svm.py {#class} 
     

   • For DWT+PCA+SVM (ours)

     python train_dwt_pca_svm.py {#class} 
     

To Test the Proving and Verification Time of Our Scheme

1. From the project root directory (i.e., 2023.2.78/src), build the rust codes

   cargo build
   

   If you see a bunch of errors due to the private modules in libspartan, you need to go to spartan libary and change such modules to public (by adding prefix pub). We will provide the updated library, where we make these changes soon.

2. Run the example on ECG dataset

   cargo run main.rs
   

3. The output from the terminal should look like as below

   Proof generated!
   Proving time:
   3
   proof verification successful!
   Verification time:
   171
   

Citing

If the code is found useful, we would be appreciated if our paper can be cited with the following bibtex format

@inproceedings{wang2023ezDPS,
  title={ezDPS: An Efficient and Zero-Knowledge Machine Learning Inference Pipeline},
  author={Haodi, Wang and Hoang, Thang},
  booktitle={Proceedings on Privacy Enhancing Technologies},
  issue={2},
  pages={430--448},
  year={2023}
}

For any inquiries, bugs, and assistance on building and running the code, please contact me at whd@mail.bnu.edu.cn