"Artificial intelligence, or AI for short, is characterized by the fact that it processes large amounts of data quickly and recognizes patterns and trends in it that are not immediately apparent to humans," explains Professor Dominik Heider, spokesperson for the LOEWE MOSLA focus area. This means AI works better and better the more data the system is provided with. With a new approach developed by Marburg researchers led by Heider, they want to ensure that more institutions share their data.
Data from various sources is essential, especially for the powerful use of AI systems in medicine. However, there are reasons, such as data protection or concerns about "data theft" by competitors, that prevent institutions from sharing it. With the goal of encouraging companies to share while ensuring a fair balance between contributions and benefits for all parties, the research group has developed a new approach.The party that contributes the most data should also get the best final result. Whereas all the other parties would at least receive models that were better than if they had only trained them using local data, explains Heider's colleague Mohammad Tajabadi. That way, he says, there are ultimately benefits for all parties.
"Organizations with more resources are more willing to collaborate with other parties for a shared learning task because the payoff is fair. But those with fewer resources also benefit from collaborating with other parties; therefore, they are likely to participate as well" Dominik Heider sums up.
The results of the study were published in the journal Jorunal of Medical Internet Research.