@@ -9,12 +9,19 @@ This folder contains code for Porthos - a semi-honest 3 party secure computation
# Running the protocol
- First setup Eigen library, used for fast matrix multiplication by Porthos, by running `./setup-eigen.sh`.
- Currently the codebase contains precompiled code for the following 3 neural networks: ResNet-50, DenseNet-121 and SqueezeNet for ImageNet, checked into the following folder: `./src/example_neural_nets`. Toggle the flag in `./src/example_neural_nets/network_config.h` to switch the network which runs. Note that if there is more than one network flag uncommented (meaning ON) or if there is already a main file in src, the compilation will error out saying multiple declarations of main function.
- To compile use `make clean && make -j`.
- To run for example the ResNet-50 code, use the following commands:
`./party1.sh < ../Athos/Networks/ResNet/ResNet_weights.inp`, and
`./party2.sh`.
- Currently the codebase contains precompiled code for the following 3 neural networks: ResNet-50, DenseNet-121 and SqueezeNet for ImageNet, checked into the following folder: `./src/example_neural_nets`.
- To compile, do the following (from Porthos root):
```
cd src/
mkdir build
cd build
cmake ..
make -j
```
- To run the MPC code for SqueezeNet, ResNet50 or DenseNet121, from Porthos root, use the following commands:
`./party1.sh [SqNetImgNet/ResNet50/DenseNet121] < ../Athos/Networks/ResNet/ResNet_weights.inp`, and
`./party2.sh [SqNetImgNet/ResNet50/DenseNet121]`.
The above commands make use of fixed-point input files generated from Athos. Please refer to the `README.md` of Athos for instructions on how to generate the same. Also, note that in the scenario of secure inference, `party0.sh` represents the client, which inputs the image, `party1.sh` represents the server, which inputs the model and `party2.sh` represents the helper party which doesn't have any input. The output is learned by `party0`, which represents the client.
