v.0.1.0

This project aims to provide Deep Learning with a privacy preserving computation backend. The privacy preserving computation is based on homomorphic encryption (HE). Homomorphic encryption allows for computation on encrypted data, get an encrypted result without the need to decrypt the data for any of the computation. These properties come with a number of restrictions and things to look out for. We aim to provide a library that provides privacy preserving deep learning for people with a deep learning background without needing to dig into HE, aswell as the abiltity to easily integrate different HE libraries as computation backends.

Out system currently assumes that the image data format is channels first. This means the contains the inputs are of the following form [batch, channel, y, x].

In Keras 2.0.6 the Flatten layer became aware of the image data format. In prior versions it was always assuming channels last. Our implementation does not suport channels first flattening in the same way that Keras does. So make sure that your Flatten layers contains the correct data format like this:

Flatten(data_format="channels_last")

The last known version of HELib supported is: ac0308715e5ae6bf5e750e8701e736d855550fc8 To obtain it use:

git checkout ac0308715e5ae6bf5e750e8701e736d855550fc8

There are issues witht he new HELib build system. It produces a way slower library. See the build section how to use the old system.

So far we support the following types of layers:

• 2D Convolutional
• Fully Connected
• Fully Connected Recurrent Layer

We support a number of activation functions:

• Linear
• Square
• ReLU ( not useable with the HE backend )
• Polynomials

We only support inference. Training is not supported and needs to be done with other tools/frameworks. As of now there is no automated way of importing pretrained models. The current suggested workflow is:

1. Train model using keras
2. Use provided tools to extract weights
3. Define model in C++ and import weights Models that are supposed to used with our code may only use these layers and activation functions.

Currently we only support HELib library as a computational backend (other than plaintext).

HE places a number of constraints and limitations on the type of computation that can be performed. This is supposed to be a list of the main points to be aware off. It is not meant to be comprehenisve.

• No Division Division of ciphertexts is not supported
• No comparison There is no comparison between ciphertext, e.g. no max, min, etc. This means we can not use ReLU, MaxPooling, etc.
• Limited number of computations Every computation performed on a ciphertext adds some additional noise to the ciphertext. If the noise exceeds some threshold we can not decrypt it correctly anymore. Therefore we can not run an arbitrary number of computations
• Limited support of Activation functions, Layers Due the constraints mentioned above we can not use every activation function or layer that we normally can. We suggest using poynomials as activation functions.

C++:

• HELib
• boost 1.67

python:

• keras
• tenserflow

At the moment only Linux is supported. We have run it succesfully on Ubuntu 16.04,
Ubuntu 17.10, Ubuntu 18.10 and Ubunutu 19.04. Other versions might work as well. Our project is based on HELib which needs to be built.

The build instructions have changed. There are performance issue with the libary file that is produced by the new HElib build system. Use the legacy system for now.

1. Follow the instructions here: https://github.com/shaih/HElib/blob/master/OLD_INSTALL.txt When installling gmp make sure to include the c++ interface by using ./configure --enable-cxx
2. We expect to find the HElib headers in the systems include path. Typically /usr/local/include. To make sure they can be found there create a directory called helib in /usr/local/include and copy all the .h files from the HElib/src directory there.
3. After installing HELib go into the build directory
4. In the objects.mk file make sure the path to fhe.a is correct for your system
5. Call make
6. The resulting binary CNNENC needs to be copied to the project to be run correctly

1. Install HELib and its dependencies. Follow the instrutctions for Option 2 library installation.

1. Check out this project.
2. Build the library using the makefile located in Debug
3. The output is Debug/libkalypso.a

If the project build fails on the HELib include, copy the .h files from the HElib directory to your global include directory.

Some parts of the project require our python module. To install it run

pip install -e python

The libkalypso.a library can be used to build deeplearning models using the HEbackend. Be sure to add the src folder to the include path.

For tips to how get started. Checkout the examples directory.

TBD

Contact: [email protected]