OpenMined and PyTorch Join Forces to Advance Privacy-Preserving Machine Learning

Today, OpenMined and PyTorch announced a strategic partnership aimed at accelerating research and development in privacy-preserving machine learning (PPML). This collaboration brings together two powerful tools—PySyft and CrypTen—to create a comprehensive ecosystem that makes PPML accessible to the broader machine learning community.

The Partnership

The partnership leverages the complementary strengths of both organizations:

• OpenMined’s PySyft: A Python library designed for secure and private machine learning, PySyft offers a flexible, user-friendly approach to implementing secure computation and privacy-preserving techniques. It excels at integrating these technologies into practical use cases such as federated learning on mobile devices, encrypted ML as a service, and privacy-preserving data science.
• PyTorch’s CrypTen: Built on PyTorch, CrypTen provides a framework for private and secure ML that implements secure multiparty computation. It offers an ML-first approach through the CrypTensor object, which mirrors the functionality of a standard PyTorch Tensor.

Fellowship Opportunities

With PyTorch’s $250,000 investment through the RAAIS Foundation, OpenMined is offering three fellowship categories:

1. Core PySyft CrypTen Integration: Fellows will integrate CrypTen as a supported backend for encrypted computation in PySyft, combining CrypTen’s high-performance secure multi-party computation with PySyft’s differential privacy and federated learning capabilities.
2. Federated Learning on Mobile, Web, and IoT Devices: This initiative will extend PyTorch to perform federated learning across various platforms, including JavaScript, Kotlin, Swift, and Python, with PySyft coordinating backends using peer-to-peer connections.
3. Development Challenges: OpenMined will host competitions focused on improving the performance and security of the PySyft and PyGrid codebases, with cash prizes for successful participants.

The Future of Privacy-Preserving ML

This collaboration promises significant advancements in making privacy-preserving technologies more accessible:

• PySyft will use CrypTen as a backend for encrypted tensors, improving performance and expanding CrypTen’s user base.
• The integration allows each library to focus on its core competencies while benefiting from the synergistic relationship.
• PyTorch is adding cryptography-friendly features, including support for cryptographically secure random number generation.

This partnership represents a major step forward in addressing the privacy and security challenges that have hindered many machine learning applications. By making these technologies more accessible to developers without cryptography expertise, OpenMined and PyTorch are helping to create a future where privacy and machine learning can coexist and thrive.

This blog post summarizes content from the original December 6, 2019, announcement published on the PyTorch blog.