Home

Awesome

zkCNN

Introduction

This is the implementation of this paper, which is a GKR-based snark for CNN reference, containing some common CNN models such as LeNet5, vgg11 and vgg16. Currently this version doesn't add complete zero-knowledge property.

Requirement

Input Format

To run the program, the command is

# run_file is the name of executable file.
# in_file is the file containing the data and weight matrix, please refer to the details below.
# config_file is the file containing config (scale and zero-point) of the model. In this implementation,
#   we don't use this file because we compute the scale and zero-point directly from the data in each
#   layer, so it's okay to put an empty file here.
# ou_file is the prediction result of this picture.
# exp_result is the experiment results filled in the table. For the definition of the table header,
#   please refer to the file `src/global_var.hpp`.
# pic_cnt is the number of pictures to be predicted. For details please refer to the following section.

${run_file} ${in_file} ${config_file} ${ou_file} ${pic_cnt} > ${exp_result}

The format of in_file

In the current code, we only allow one picture and one matrix to be put in this file. If pic_cnt > 1, then the code will internally duplicate the data for corresponding times. Thus to test different pictures, you might need to adjust the code of loading the input.

Data Part

This part is for a picture data, a vector reshaped from its original matrix by

formula1

where formula2 is the number of channel, formula3 is the height, formula4 is the width.

Weight Part

This part is the set of parameters in the neural network, which contains

The format of config_file

Typically this is a file to record scale and zero-point for the data in each layer. However, in our current implementation, those are computed directly from the those data. Therefore, you can just leave it blank.

Experiment Script

Clone the repo

To run the code, make sure you clone with

git clone --recurse-submodules git@github.com:TAMUCrypto/zkCNN.git

since the polynomial commitment is included as a submodule.

Run a demo of LeNet5

The script to run LeNet5 model (please run the script in script/ directory).

./demo_lenet.sh

Run a demo of vgg11

The script to run vgg11 model (please run the script in script/ directory).

./demo_vgg.sh

Polynomial Commitment

Here we implement a hyrax polynomial commitment based on BLS12-381 elliptic curve. It is a submodule and someone who is interested can refer to this repo hyrax-bls12-381.

Reference