crypten

CrypTen is an open-source framework for privacy-preserving machine learning that enables secure multi-party computation (MPC) for neural networks and other algorithms. It is designed to let multiple parties jointly train or run inference on machine-learning models while keeping each party’s raw data private. Computations are performed on secret shares rather than plaintext data, and only the final results are disclosed to the designated parties.

CrypTen is implemented in Python and built on top of PyTorch, exposing a PyTorch-like API that makes it easier for researchers and practitioners to port existing models to a privacy-preserving setting. Users define models and training routines much as they would in PyTorch; CrypTen handles the distribution of data shares, the orchestration of secure computation across parties, and the application of cryptographic protocols to arithmetic and basic neural-network operations.

The library supports common ML primitives such as linear and convolutional layers, activations, and loss functions, with secure protocols used to operate on shared data. It focuses on practical performance for ML workloads and is intended for research and experimentation in privacy-preserving ML rather than production-grade deployment in all settings.

CrypTen is open source and maintained by a community of researchers. It is typically used for collaborative ML studies across institutions where data cannot be shared directly, such as in healthcare or finance, to enable joint learning while protecting participant privacy. Documentation and tutorials accompany the project on its repository.