Top science team of UMCG attends virtual Bayesian Networks training by GAIn®

By Ilya Petoukhov | Program Manager at GAIn® – The Global AI network


Earlier this week, we hosted a training in one of our most promising new topics: Bayesian Networks – a method which shows how different variables together cause an event to occur. The strengths of the method lie in the fact that it can estimate and show the dependencies between all variables in the model and that prior field knowledge by a research team can be formally included in the model. Despite the method still being relatively new in a business context, it has already been extensively used in (medical) research and its great potential attracts professionals from diverse industries to learn more. This week’s training was unique, as, in addition to Data Scientists, there were three researchers and four professors attending from the University Medical Center of Groningen (UMCG). While some were already familiar with the basics of Bayesian Networks, they all wanted to broaden their knowledge of the topic to leverage the advantages of this algorithm for their own field, mainly in discovering how our genetics lead to lung diseases. Teaching Bayesian Networks to this talented group of participants made us learn a lot ourselves too, about how to further improve and expand this training.



Thanks to a virtual setting, our participants from across the Netherlands could be trained by one of our GAIn experts in Israel. We asked the participants what they thought of the virtual training and how they would apply Bayesian Networks in their own work. The professors from UMCG were asked how it felt to be a student again. Read on to find out what they said!

‘’How was your overall experience
of the training?’’

Gerard Koppelman: It was a well-structured training, taking us from basics to complex applications. Great to see how interactive you can make a Zoom session.


Martijn Nawijn: The training was very hands-on. Performing the analyses yourself really helped understand the power and limitations of the models.

Wim Timens: Excellent teachers! They really improved my insight into all the possibilities of Bayesian Networks, which I didn’t know much of before. Examples were very clear. For me not knowing much of R was a challenge but others gave good instructions, and it motivated me to upskill and learn R. I enjoyed the training!


Victor Guryev: Good mix of theory and practice and really nice training set you’ve used.

Bayesian networks – from basics to complex applications

The 1-day training covered the full process of Bayesian Networks – from model creation to analysis. It was explained that where traditional techniques mostly focus on achieving high prediction performance, Bayesian Networks excel at showing why something happens, while incorporating expert knowledge in the model. Additionally, the visual way in which the network is drawn, easily shares this insight with your audience. These advantages have made Bayesian Networks a popular modelling choice in medical research, but it has also started to gain traction in the business world.


Applying Bayesian Networks in practice

Several use cases of Bayesian Networks were presented during the training, such as using it for understanding causes of disease, targeting terrorists via call records and performing preventive maintenance on trains. However, the participants had even more ideas on innovative ways of applying this methodology, and many said they were already doing so or planning to in the near future. There were thoughts around using Bayesian Networks for different types of process mining  and to better understand the maintenance needs for telecommunication towers, so they can be tackled at the source. It was even suggested that we now would be able to predict the probability of Groningen beating Ajax, given that Arjen Robben is joining the Groningen team. The professors from UMCG, who already had some experience working with Bayesian Networks on a project about allergies in children, said they would like to revisit this case and apply the newly learned algorithms, such as Targeted BN’s, to get more insight into the relations within allergy networks.



Student for a day

For the four UMCG professors, who have all been teaching for 20 to 35 years, it was an extraordinary day. Despite their normally full schedules, they all managed to free up an entire day to go back to being a student. And boy, they still knew how to do it! Asking a lot of sharp questions and engaging effortlessly with both basic and advanced content, it was a wonderful group to teach. We asked them how it felt for them to attend the training as a professor and step into the students’ shoes.

How was it as a professor to be a student for a day?

Wim Timens: On the one hand I found it challenging, on other hand very much appealing as the training made it very clear how we could use Bayesian Networks in our research. I was glad I prepared somewhat before, otherwise the challenge might have been very big. But learning new things is always fun!


Martijn Nawijn: The training was very useful for all participants, ranging from PhD students to full professors.

Gerard Koppelman: Never stop learning. Great didactic quality, so a pleasure to attend this training. Also, can you provide me with continuing education points?


Victor Guryev: Nice to be a student again, I have not had exams and assignments for ages!

Never stop learning

Gerard Koppelman from UMCG put it well – never stop learning! And we never do. This is why all of our faculty is constantly involved in learning the newest AI techniques, either through a three-year development program or a continuous learning program for the more experienced colleagues. We’re excited to keep diving deeper in to the broad array of AI algorithms available to us and and to keep sharing these findings with you. Our open course calendar offers the possibility to learn about the newest algorithms and best practices, such as how to detect anomalies in data (1&2 Sept), how to deal with datasets containing many variables (20&21 Oct) and advance your skills past traditional statistical methods to make stronger causal claims (1&2 Dec).