Secure and Private AI
Learn how to extend PyTorch with the tools necessary to train AI models that preserve user privacy.
About this Course
What’s the earliest we can predict cancer survival rates, and what schools do the best job of educating children? You can only answer these questions with very rare access to private and personal data, but access to this personal data requires that you master methods for the principled protection of user privacy. While not all privacy use cases have been solved, the last few years have seen great strides in privacy-preserving technologies.
This free course will introduce you to three cutting-edge technologies for privacy-preserving AI: Federated Learning, Differential Privacy, and Encrypted Computation. You will learn how to use the newest privacy-preserving technologies, such as OpenMined's PySyft. PySyft extends Deep Learning tools—such as PyTorch—with the cryptographic and distributed technologies necessary to safely and securely train AI models on distributed private data.
We encourage you to enter the Secure and Private AI Scholarship Challenge from Facebook to both take the course and have a chance to win a scholarship for the Deep Learning or Computer Vision Nanodegree programs.
Approx. 2 months
Included in Product
Rich Learning Content
Taught by Industry Pros
Leader of OpenMined, Research Scientist at DeepMind Oxford, PhD Student
What You Will Learn
- Learn the mathematical definition of privacy
- Train AI models in PyTorch to learn public information from within private datasets
- Train on data that is highly distributed across multiple organizations and data centers using PyTorch and PySyft
- Aggregate gradients using a "trusted aggregator"
Prerequisites and Requirements
To get the most out of your experience in this course, we recommend the following:
- Beginner-level skills in Deep Learning or Machine Learning
- Beginner-level skills in at least one Deep Learning framework (such as PyTorch)
- Beginner-level skills in Python
No background in cryptography or advanced mathematics is required.
See the Technology Requirements for using Udacity.
Why Take This Course
A data scientist can only use AI to solve problems if they have enough training data. Whether you're at a startup or an enterprise, the most important and valuable problems are problems about people. Solving these problems using AI means having access to a large amount of private and sensitive data.
Want to predict cancer in medical scans? If you're using traditional Deep Learning tools, this means persuading someone to send you a copy of a sensitive dataset. In many cases, this is either a non-starter or it will severely limit the amount of data you're allowed to see.
In this course, learn how to apply Deep Learning to private data while maintaining users' privacy, giving you the ability to train on more data in a privacy-preserving manner so that you can tackle more difficult problems and create smarter, more effective AI models, while also being socially responsible.
What do I get?
- Instructor videos
- Learn by doing exercises
- Taught by industry professionals