NEWS Soon, our digital devices will be even better at adapting to our needs – without compromising our personal privacy. This is the conclusion of Sourasekhar Banerjee in a new doctoral thesis in computer science at Umeå University.
Sourasekhar Banerjee during the nailing ceremony at Umeå University Library.
ImageVictoria SkeidsvollWhen we use computers and mobile phones to browse the web and use apps, we share large amounts of data. For example, information is collected about our location, what we click on, and how long we spend on various websites. These data points can be used to map our preferences and online behavior. Recently, the debate about user data has focused on how this impacts our personal privacy.
In his thesis, Sourasekhar Banerjee has explored how digital devices can be made more efficient and personalized while simultaneously protecting our privacy. The research focuses on a technique called "federal learning" – or collaborative learning – where multiple devices work together without directly sharing data with each other. Instead, each device holds different pieces of information, and together they form a complete picture. By connecting multiple devices in this way, it becomes possible to control what data is shared and when. This enables personalized digital services for the user without the need to share sensitive information with third parties.
”Imagine healthcare providers offering personalized medical advice without compromising patient privacy, or financial apps giving customized investment advice without revealing sensitive financial information," says Sourasekhar Banerjee.
Banerjee emphasizes that companies in healthcare, finance, and digital services can greatly benefit from these techniques to offer smart and privacy-protected AI solutions. The technology can also be used in everyday apps, such as photo applications on mobile phones, where images can be analyzed without the personal data ever leaving the device.
On Monday, September 23, Sourasekhar Banerjee from the Department of Computer Science at Umeå University will defend his doctoral thesis titled Advancing Federated Learning: Algorithms and Use-Cases. The defense will take place at 13:00 in Lecture Hall HUM.D.210, Humanisthuset. The opponent is Salman Toor, Associate Professor of Scientific Computing at Uppsala University.