MENLO PARK — In a new initiative to enhance privacy in artificial intelligence and machine learning (ML), Facebook AI has joined forces with OpenMined, an open-source community that specializes in privacy-focused AI development. Together, they are launching an educational series titled The Private AI Series, designed to teach developers about privacy-preserving machine learning (PPML) techniques using the PyTorch framework.

As ML models increasingly work with sensitive data, it is essential to maintain stringent data privacy standards. The goal is not only to secure data from potential hackers but also to ensure that the models, the information they generate, and the data they utilize remain inaccessible to unauthorized users.

The Private AI Series is structured to cater to developers at all skill levels, from beginners to advanced practitioners. The courses will provide a comprehensive introduction to privacy-enhancing technologies (PETs), alongside hands-on experience in developing secure ML models. The initiative aims to raise awareness about PPML concepts while encouraging the practical adoption of key privacy-focused technologies. Participants will use PyTorch and PySyft, a Python library specifically designed for privacy-preserving machine learning.

The series consists of four courses, each targeting different aspects of privacy and machine learning:

1. Awareness:
This foundational course introduces privacy-enhancing technology (PET) and its growing influence across industries. Participants will explore the importance of privacy, how PET is reshaping business operations, and the opportunities it presents. By the end, students will have a solid grasp of core privacy technologies and how they are set to disrupt traditional business models.

2. Foundation:
This course dives deeper into the technical side, teaching students how to design, evaluate, and implement systems that utilize PETs. Topics will include federated learning, differential privacy, homomorphic encryption, and cryptographic techniques. Graduates of this course will be equipped to read and apply technical papers on PETs, allowing them to build privacy-focused systems.

3. Cross-Enterprise Statistics and Federated Learning:
Focusing on real-world applications, this course covers how to perform analysis across organizations using PyTorch and PySyft. Students will learn techniques such as searching for unseen data, executing remote statistical models, and working with datasets across multiple enterprises while ensuring privacy.

4. Federated Statistics and Learning on Web and Mobile:
The final course centers on web and mobile applications. Developers will learn how to build apps that store data on user devices and train models through federated learning. Using PyTorch, participants will perform distributed database queries and build privacy-focused applications for platforms such as Android, iOS, and React.js.

The Private AI Series will kick off with the Awareness course in January 2021. Developers interested in enhancing their skills in privacy-preserving technologies can register at courses.openmined.org for more details.

Through this collaboration, Facebook AI and OpenMined aim to cultivate a new generation of developers proficient in privacy-preserving machine learning techniques, ensuring that AI advancements continue to prioritize data security and user privacy.