Applying privacy-preserving ML techniques

We have covered many approaches to help us protect data privacy. None of these is a silver bullet. How we apply them will depend on the context of the application, and requires a balancing act between privacy and data utility and a multi-layered defense-in-depth approach:

• Risk and use case assessment: Understanding the specific risks in a given use case, including data governance and compliance requirements, is essential in choosing techniques such as anonymization on sensitive data, how we use them, and where we apply them.
• Threat modeling: Identifying potential threats helps us understand these risks better. This is essential in evaluating privacy attacks that rely on unforeseen data linkage.
• Data minimization. Reducing the amount of sensitive data used minimizes the attack surface and risk. Using DP is an essential tool to help minimize data linking.
• Balancing data utility: Ensuring data remains useful for its purpose...