Overview of AI Privacy

1. Overview of AI Privacy

Welcome to AI Privacy. This module consists of eight lessons. Today, you will learn to define privacy in AI, discover some best practices on privacy, describe the types of security behind privacy, explore techniques and tools for data and model security for privacy and explain how to apply security best practices for generative AI models in Google Cloud. Let's start with an overview of AI privacy. Why do you need to take privacy into consideration? There are several reasons. This includes not only respecting the legal and regulatory requirements, but also considering social norms and typical individual expectations. A sensitive attribute is a human attribute that may be given special consideration for legal, ethical, social, or personal reasons. This includes, but is not limited to, personal identifiable information such as full names, date of birth, address and phone number, social data such as ethnicity, religion, sexual orientation, and political affiliation, financial data such as credit card numbers, income, and tax records, health and medical data such as diagnosis, prescriptions, and genetic data, geolocation data such as tracking data from mobile devices, biometric data such as facial recognition, voice, and fingerprints, user authentication data such as usernames, passwords, security questions and answers. Legal data, including intellectual property and trade secrets. There may be enormous benefits to building a model that operates on sensitive data such as an AI system trained on electronic health records to predict an individual's risk of chronic diseases, then deployed to make personalized health recommendations. However, it is always essential to consider the potential privacy implications in using sensitive data. How can you ensure you are building a secure machine learning based system? There are several aspects to consider. Security in training data. Since machine learning is always trained on a dataset, the protection of sensitive and confidential data used for AI systems is critical. Security in training process. You can address security by adopting recommended practices for secure machine learning training like federated learning and differentially private stochastic gradient descent or DP-SGD. Implementing privacy preserving techniques across hardware, software, communication channels, and infrastructure is crucial for comprehensive AI system security. We'll look at data security in detail first, then discuss the other two aspects.

2. Let's practice!

Create Your Free Account

By continuing, you accept our Terms of Use, our Privacy Policy and that your data is stored in the USA.