ITU receives two Danish Data Science Academy Fellowships
Each year, the DDSA awards a total of 10 PhDs, and 6 postdocs. This year, ITU has secured two – Nils Grünefeld who will undertake a PhD in Machine Learning and Natural Language Processing, and Ola Rønning will begin a postdoc project in Probabilistic Programming.
Researchartificial intelligencealgorithmsdata science
Written 26 June, 2025 07:54 by Theis Duelund Jensen
The prestigious Danish Data Science Academy Fellowships have been announced and this year ITU has received no less than two fellowships.
Working with Associate Professor Christian Hardmeier of the Data Science section, Nils Grünefeld will undertake a PhD focused on optimising ways of evaluating the credibility and accuracy of large language model output. In the Software Engineering section, Professor Andrzej Wasowski will be working with Postdoc Ola Rønning on a project aiming to improve the localization and positioning abilities of robots (particularly those operating under water).
Helping lost robots
In his postdoc project, Ola Rønning will extend Stein mixture inference—a flexible, uncertainty-aware algorithm—from offline batch data to real-time streams. The main focus is on robotics, where the researcher will apply Stein mixture filtering to mobile navigation and, for example, enable a submarine robot to recognize when it is lost and trigger emergency protocols. Uncertainty-aware systems can quantify risk, which is crucial for high-stakes, real-time decisions in fields like medicine, industry, and finance.
“It is not often you can find a person with the sufficient theoretical insight to address such problems who is willing to work on applications, and it is not often you find a robotics engineer able to improve basic statistics of pose estimation. I really look forward to working with Ola on this unique combination of challenges,” says Andrzej Wasowski who is head of the Software Quality Research (SQUARE) at ITU.
“I’m thrilled to have been awarded a Danish Data Science Academy Fellowship and to join the Software Quality Research (SQUARE) group at ITU – a unique opportunity to become deeply embedded in the Danish data science community. SQUARE’s combination of theoretical rigor and real-world impact, especially its work in marine robotics and contributions to the ROS ecosystem, makes it an ideal environment for advancing my research on uncertainty-aware robotics,” says Ola Rønning.
AI and reliability
AI chatbots and similar tools based on large language models have an enormous impact on society, and in record time we have become accustomed to using them in all manner of settings – from everyday tasks in our personal lives to high-stakes decision making in professional settings. But even if the output of an LLM sounds very convincing, it’s not necessarily accurate. Predicting to what extent we can trust any specific LLM output is a difficult task, not least because of the sheer size of the models, which render many established methods impractical.
This is where Nils Grünefeld’s PhD project which is supervised by Associate Professor Christian Hardmeier from ITU together with Associate Professor Jes Frellsen from DTU comes into play. The goal of the project is to develop a novel and more efficient approach based on specific properties of human language, which will enable the researchers to determine how far LLM output can be trusted thus enabling the use of LLM agents in high-stakes situations with much greater confidence.
“When an AI is prompted it provides an answer, but behind the scenes there is always a degree of uncertainty about that answer. There is always a set of likely answers, each with a probability attached to it. The AI only gives you the single most likely answer, but it makes a big difference whether an AI is 95 or 55 % certain about an answer, especially in high-risk fields like the medical field,” says Nils Grünefeld.
The results will be tested in a healthcare context in collaboration with the VIRTU research group at Region Hovedstadens Psykiatri (Louise Glenthøj and Benjamin Arnfred), and it also includes a collaboration with the leading natural language processing group at the University of Edinburgh (Ivan Titov).
{Box}
Theis Duelund Jensen, Press Officer, phone +45 2555 0447, email