Dimitris N. Chorafas Foundation Award 2021 – Sai Praneeth Karimireddy
Optimization algorithms for collaborative machine learning
Thesis director: Prof. M. Jaggi
For ground-breaking work on collaborative learning, in particular for efficient and robust federated learning, and for advancing communication-efficient distributed training of machine learning models.
A traditional machine learning pipeline involves collecting massive amounts of data centrally on a server and training models to fit this data. However, there are increasing concerns about privacy and security, as well as legislation preventing sharing of data. Thus, there is a strong need to look beyond such traditional centralized approaches. This work instead develops an alternative paradigm for machine learning called collaborative learning. We show how a network of data holders (such as hospitals, clinics, or research centers) can combine their locally trained machine learning models together in an accurate, secure, and private manner. This allows them to directly collaborate together to train models on their joint data, without transmitting any raw data to each other.