New research project to find a more inclusive way to develop algorithms
Associate Professor Veronika Cheplygina has received a Novo Nordisk Data Science Investigator Grant of almost DKK 11 million. The grant will fund research on how more inclusive teaching and research environments may lead to better algorithms for medical imaging.
Veronika CheplyginaResearchalgorithmsgrantshealthartificial intelligence
Written 10 February, 2025 08:00
During her academic career, Veronika Cheplygina has never chosen the beaten path. As a bachelor student, she was skeptical of the practice of turning class assignments in programming into coding competitions. She noted that the practice in fact had a tendency to exclude many students not eager to compete. This, among other experiences in academia, sparked her interest in inclusive ways of teaching computer science and made her question if competitions are the best way to drive scientific innovation.
Competition and competitiveness might drive away underrepresented researchers. I believe, data science and the developments in the field will benefit from more diverse groups of data scientists
Associate Professor Veronika Cheplygina
“When you develop an algorithm in a competition, usually you develop it to succeed on one specific evaluation metric, for example how many cancer patients were found. Therefore, there may be a lot of blind spots, for example whether the algorithm is effective for different ages and genders of patients. Also, aspects like the carbon footprint of training the algorithms is often not considered. Often, we forget to look at how the algorithms will be used in the real world, for example will a hospital in a rural area be able to run this algorithm?” says Associate Professor at ITU, Veronika Cheplygina, “I want to find out if competitions where algorithms need to succeed on multiple evaluation metrics, lead to both better algorithms, and more diverse teams participating.”
Competitions are common in data science – both at conferences and now increasingly in classrooms. Students or conference participants compete in developing the best new algorithm which is optimized for one specific evaluation metric. Veronika Cheplygina’s research revolves around medical imaging, and in this field these algorithms may be effective for specific purposes while failing others.
Competing in all aspects
Data science is a competitive field. Competitions in algorithm development tools are not alone in driving it.
“There is a big focus on publications, funding, and developing novelty algorithms, and I believe that is not ideal for the field. Competition and competitiveness might drive away underrepresented researchers, such as women or researchers at institutions which might not have the resources to train large algorithms. I believe, data science and the developments in the field will benefit from more diverse groups of data scientists,” says Veronika Cheplygina.
As part of the project, Veronika Cheplygina and her team want to measure the quality of data used for evaluating the developed algorithms, and how diverse the developed algorithms are – since more diverse algorithms with different strengths and weaknesses could be combined into a more effective algorithm. She also wants to ask participants about what they thought of the competition process, such as why did they join (or not join) the competition. This has not been researched before.
“This does not necessarily mean that competitions are bad and should not be used. But we need to find out if there is a way to do the competitions that is more inclusive for everyone,” says Veronika Cheplygina.
She hopes that the research will not only come out as papers in journals, but it will also be usable in practice.
“We will also talk to students and get their perspective on the competitions. I hope we can develop some tools and proposals on how the format could be different to improve all these things. If we have a competition platform that people use by the end of this, and we see more willingness from diverse people to participate in competitions and ultimately staying in this field, I think we have succeeded,” says Veronika Cheplygina.
The project is entitled “CHEeTAh: CHallenges of Evaluating Teams and Algorithms” and will run for five years.
Theis Duelund Jensen, Press Officer, phone +45 2555 0447, email thej@itu.dk