CrypTen is a research framework developed by Facebook Research for privacy-preserving machine learning built directly on top of PyTorch. It provides a secure and intuitive environment for performing computations on encrypted data using Secure Multiparty Computation (SMPC). Designed to make secure computation accessible to machine learning practitioners, CrypTen introduces a CrypTensor object that behaves like a regular PyTorch tensor, allowing users to seamlessly apply automatic differentiation and neural network operations. Its design mirrors PyTorch’s modular and library-based structure, enabling flexible experimentation, debugging, and model development. The framework supports both encryption and decryption of tensors and operations such as addition and multiplication over encrypted values. Although not yet production-ready, CrypTen focuses on advancing real-world secure ML applications, such as training and inference over private datasets, without exposing sensitive data.
Features
- Implements privacy-preserving machine learning using Secure Multiparty Computation
- Provides CrypTensor, a PyTorch-like encrypted tensor supporting autograd and modules
- Enables encrypted model training and inference with minimal code changes
- Offers rich tutorials and examples for encrypted neural networks and models
- Supports GPU acceleration for efficient encrypted computation
- Focused on research use cases and extensible for experimental privacy applications
Project Samples
