Protecting Patient Privacy in AI for Responsible Sharing & Deployment

Course description

This course introduces the privacy risks and vulnerabilities which exist in training AI models on sensitive healthcare datasets, and releasing them from trusted research environments. You will learn about these different privacy-preserving techniques and how they can be implemented given the type of data and models being used, as well as how it is being deployed or shared. You will also learn how to effectively evaluate your models to ensure they are private and safe.

 

Learning mode

Online, pre-recorded videos with online reading material

 

Who should attend

Researchers who develop AI models should take this course to learn how their models could be compromising privacy and what they can do to ensure the models they develop can be responsibly deployed.

 

Aims and objectives

Objectives:

• Understand the privacy risks in AI models:

• • Overfitting
  • Membership Inference Attacks
  • Attribute Inference Attacks
  • Reconstruction Attacks
  • Challenges with Complex Data Types

• Learn how to mitigate these risks through:

• • Data minimisation
  • Synthetic Data
  • Differential Privacy
  • Homomorphic Encryption

• Be able to evaluate privacy in AI models and run attacks to assess them

 

Skills area and competencies

• Data Privacy and Stewardship
  • Ensuring the Protection of Personal and Sensitive Data

 

Pre-requisite

Anyone taking this course should have experience of developing AI models and python programming

 

Learning outcomes

By the end of this course you will learn how to comply with best practices in developing responsible AI and ensuring that models are developed in line with data protection regulations.

 

Transferrable skills

• Understanding of data protection
• AI governance

Assessment

Multiple choice quiz to self-check understanding

 

Delivered in partnership with