PhD defence Afsana Khan

Keywords: Federated Learning, Data Privacy, Distributed Data, Responsible AI

"Unlocking Value of Data with Vertical Federated Learning"

Data that could help organisations detect risks, improve services, or make better decisions is often spread across different companies and institutions. However, bringing all this information together in one place is rarely straightforward. The data may contain personal, sensitive, or commercially valuable details, and is usually protected by privacy, security, legal, or business restrictions. 

Federated learning offers a solution. This privacy-preserving approach to machine learning enables organisations to jointly develop a shared model for prediction or decision support, without ever moving their data from their own systems. 

Afsana Khan’s research focused on vertical federated learning, a method used when the information needed for a decision is split across organisations in complementary ways. In these cases, no single organisation holds all the necessary data, but collaboration can lead to better, more informed decisions. Her research explored practical challenges: which organisations should participate, how to minimise unnecessary information sharing, how to fairly reward contributors, and how to handle the complexities of real-world data distribution.

Click here for the live stream.