Patrick Denantes Memorial Prize 2022

In our hyper-digital world, data is key which means data privacy continues to be a hot topic. Currently, for data to be useful, for example in training artificial intelligence models a large central entity collects and stores it from a lot of users. From then on, the users have no control or ownership over their data.

Collaborative AI envisions an alternative - the users themselves collaborate with each other to train machine learning models without transmitting any raw data. This way, they can retain full control and ownership over their data, while still deriving utility from it. Making this a reality however still faces significant challenges, and requires building "intelligence infrastructure" which is the focus of Sai Praneeth Karimireddy’s research.

“This issue inspired the work of my thesis. Consider, for example, the marvel of modern financial infrastructure. We store our money in banks and have full control over it. However, behind the scenes, it is also being harnessed to provide loans and investments. I envision a similar infrastructure for our data. We should be able to form data banks or data co-ops where we harness our collective data in a privacy preserving manner. The monetary value from this is then distributed back, just as we get interest for our money deposited in banks,” said Karimireddy, now a Swiss National Science Foundation postdoc researcher at UC Berkeley.

While this is a long-term vision, the immediate application today is enabling collaboration between large organizations with highly sensitive data, for example, hospitals or pharmaceutical companies that can collaborate to train AI models on their combined medical data without revealing any sensitive patient information.

“I am currently working with a few medical organizations, as well as in talks with cancer registrars to enable such collaborative AI. I am extremely excited and motivated to be able to translate some of my research to the real world and I feel incredibly honored to have won this award. It was a great surprise and I was giddy the entire week that I received the news,” Karimireddy continued.

He is enjoying his new role at UC Berkeley and the inter-disciplinary research that he is able to undertake, while being grateful for the research foundation he received at EPFL in the Machine Learning and Optimization Laboratory (MLO).

“I have grown so much in my approach to research. EPFL and my advisor Martin Jaggi gave me the luxury to explore and learn, instead of focusing narrowly on publications. I hope to continue this approach of taking my time to carefully pick and formulate the research questions before rushing to answer them. The most valuable lesson was the importance of taking care of myself, and to not take my work too seriously. A research career is long. There is no rush and absolutely no point in burning out.”

“It’s always a pleasure to see the hard work of students recognized and Praneeth’s focus on collaborative AI alternatives to how data is currently used is so important when it comes to data privacy for us all. I look forward to following his career and research path to see some of his big ideas come to life,” said Jaggi, Head of the Machine Learning and Optimization Laboratory.

This Memorial Prize was created in 2010 to honor the memory of Patrick Denantes, a doctoral student in Communication Systems at EPFL who tragically died in a climbing accident in 2009. It is awarded annually to the author of an outstanding doctoral thesis from the School of Computer and Communication Sciences. The prize is awarded by a jury and presented to the laureate at the School’s end-of-year event. Financial sponsorship is provided by the Denantes family and the Nokia Research Center.