Personal AI solutions challenge privacy

When Alexander Galozy wrote his licentiate thesis, it was part of the iMedA project, where the research group developed a mobile application with customised reminders and personalised information. To assist patients in adhering to their treatment and medication plans, the mobile application needs more information to understand the patient group better. Therefore, large amounts of existing electronic health records (EHR) are analysed using machine learning models. Since the licentiate thesis, Alexander Galozy has, among other things, worked on improving the learning speed of the algorithms.

“We have tested increasing the speed at which our algorithms more efficiently utilise previously collected data and intelligently analyse patterns. This has reduced the number of interactions required to offer customised solutions in the mobile application. This not only enhances user engagement but also strengthens the ability to provide patients with more effective and personalised interventions,” says Alexander Galozy.

Imbalance between performance and privacy

With his fellow researchers, Alexander Galozy has also investigated privacy concerns regarding user data by analysing how effective simple noise-additive schemes are at preserving user data privacy in a recommender system. A recommender system is an IT-support system that uses AI algorithms on big data to make suggestions and recommendations to the end-user, for example, a doctor. By adding noise, the researchers hoped to mask and preserve private data about individual patients.

“The recommender system is not unlike something you would use in mobile health (mHealth) setting to provide digital interventions”, Alexander Galozy says. He continues, “We have found that these noise-additive schemes, unfortunately, are insufficient for preserving privacy or destroying too much of the information. There is an interesting avenue for further research to find a better performance-privacy trade-off.”

AI solutions for personalised care

Alexander Galozy believes that mHealth will become more and more important in the future as the population ages. He hopes his research will make it easier to apply probability methods when developing algorithms and applications, more specifically Bandit methods, that can help make suggestions, estimations and choices. Alexander Galozy thinks that this will be useful not only in the mHealth domain but also in other domains.

“It will allow patients to receive good interventions which are more tailored to their needs – right when they need it”, Alexander Galozy says. He continues, “Also, I think the topic of privacy and how it affects performance in the Bandit research is just emerging, and we do not know a lot yet. With my research, I provided a little step forward to address the issue of privacy in this domain, as it will become, and certainly is, very relevant today and in the future.”

Text: Anna-Frida Agardson

Bild överst: iStock

Bild övriga: Anna-Frida Agardson