TenSEAL

Changelog:

• PlainTensor serialization support

Changelog:

• Lazy context support: allows us to deserialize tensors and link the context later(https://github.com/OpenMined/TenSEAL/pull/197).
• CKKSTensor reshape support (https://github.com/OpenMined/TenSEAL/pull/194).
• CKKSTensor Dot operation(https://github.com/OpenMined/TenSEAL/pull/196).
• matmul API updates (https://github.com/OpenMined/TenSEAL/pull/208)
• TensorStorage with xtensor(https://github.com/OpenMined/TenSEAL/pull/198)
• Symmetric encryption support (https://github.com/OpenMined/TenSEAL/pull/210)

We introduce new Python wrappers for all low level objects (#185). Previously, the python API interacted directly with the binded low level objects. However, we now have a Python class wrapping every low level objects.
The API is still the same.

This version mainly introduce a new tensor type, the CKKSTensor, which can encrypts N-dimensional tensors, with the ability to use one axis as the batch_axis, which will be stored along ciphertext slots.

New features

• CKKSTensor (#156, #168, #169, #170, #171, #173)
• PlainTensor (#167, #170)

Bug fixes

• Race condition on multiworkers (#151)
• Fix Centos scripts (#149)

Other

• We now link protobuf statically, so the running system doesn't need the library to be installed (#177)
• We now use SEAL 3.6 (#158)
• Refactoring and library improvements (#153, #167)

Fix package build on release #147

The convolution operation introduced in this alpha release have an experimental API that should be updated in upcoming releases.

New features

• Serialization of TenSEALContext and tensor types (#101, #109)
• Threadpool support and parallel vector-matrix multiplication (#115, #124)
• Check if a key is found in a TenSEALContext (#132)
• Convolution using im2col technique (#117, #134, #138)
• Encrypted matrix multiplication with plain vector (#137, #138)
• Vector packing (#141)

Bug fixes

• make_public_context should keep the existing keys (#110)
• Better handling of transparent ct exceptions (#119)
• Check for empty vector upon encryption (#126)
• Fix CKKS and BFV vectors constructors (#133)

Other

• Tutorial 1 - Training and Evaluation of Logistic Regression on Encrypted Data (#85)
• Update docker images, docs and tags (#104, #111, #114)
• Upgrade SEAL to 3.5.6 (#105)
• Bazel build (#116)
• Automatic tests on [Linux, MacOS, Windows] x [py36, py37, py38] (#128)
• Code optimization (#142)
• Fix release workflow (#143)

• Scalar operations for CKKSVector (#86)
• Operations on vectors of different size (1 vs n) (#92)
• Negate and right arithmetic operators for CKKSVector (#93, #94)
• Encrypted vector exponentiation and polynomial evaluation with optimum circuit (#99, #100)

With all this recent changes, TenSEAL is now able to perform training of a logistic regression model, checkout the tutorial section for more information.

• Fixing package build for MacOS (#87 ) and Windows (#89)

This is the first official release of TenSEAL, so we will mainly focus on what this first version have been building so far

• Encryption/Decryption of vectors of real numbers or integers using either CKKS or BFV respectively
• Element-wise addition, substraction and multiplication of encrypted-encrypted vectors and encrypted-plain vectors
• Dot product and vector-matrix multiplication
• Complete SEAL API under tenseal.sealapi
• Support for Linux, Windows and MacOS
• Support for Python 3.6, 3.7 and 3.8
• Docker images with TenSEAL for Python 3.6, 3.7 and 3.8
• Tutorial for Getting Started
• Automation of build, release and tests of both Python and C++ code.

Thanks to all the contributors of the TenSEAL library, more specifically to @philomath213 and @bcebere for their time and efforts spent to be able able to make this first release.