Neuro-mechanistic modeling combines data-driven models with domain knowledge

Double interview with Prof. Dr. Verena Wolf and Prof. Dr. Antonio Krüger

Prof. Dr. Verena Wolf has been the head of the new research department of Neuro-mechanistic Modeling at DFKI Saarbrücken since January 1, 2023. She is a professor of computer science at Saarland University (UdS), where she holds the chair for modeling and simulation. Verena Wolf was instrumental in making computer science a compulsory subject in all types of schools in Saarland from grade 7.

Prof. Dr. Antonio Krüger has been Chairman of the Board of DFKI since November 2019, heads the DFKI research area Cognitive Assistance Systems as Scientific Director, established the Media Informatics degree program at Saarland University in 2010, and is still responsible for it today. He was a member of the Enquete Commission on Artificial Intelligence of the German Bundestag.

What is the approach of the new research department to neuro-mechanistic modeling?

Verena Wolf: Neuro-mechanistic modeling stands for research approaches in which we combine models of different types in order to use the different strengths of the model classes involved with a hybrid approach: Data-driven models based on artificial neural networks and classical mechanistic models, with which we describe mechanistic hypotheses and bring in domain knowledge.

Antonio Krüger: Neuro-mechanistic modeling is oriented towards two paradigms of AI research. At its core, it is about combining the "net-like" of artificial neural networks with the "law-like" of explicit modeling in such a way that the resulting applications can benefit from the best of both worlds. Artificial neural networks allow the identification of patterns in the respective over-complex subject areas; explicit knowledge models contribute to comprehensibility and allow robust statements about reliability.

Keyword model types, what does modeling mean in the informatics sense?

Verena Wolf: This is the mathematical description of real or imagined systems based on hypotheses about the mechanistic relationships between measurable quantities.

How does Prof. Wolf's scientific work fit into the research landscape at DFKI?

Antonio Krüger: The neuro-mechanistic approach of Verena Wolf's research is domain-open, sharpens DFKI's research portfolio, and strengthens DFKI's expertise in hybrid AI, i.e., in the combination of model-based and sub-symbolic methods.

For which application areas is this research approach particularly promising?

Verena Wolf: Wherever domain knowledge is available in an explicit way – such as in life sciences – and the generation of very large amounts of data is expensive or impossible. For example, if we want to develop new active ingredients with the help of AI, we can directly integrate existing knowledge about chemical compounds and molecular properties as explicit rules. We don't have to relearn that for every model. The central question is: How do we bring the already established knowledge into the machine learning process?

To what other domains are neuro-mechanistic models applicable?

Verena Wolf: Mechanistic and data-driven approaches can also be combined in the field of production and logistics. In many areas, we have Digital Twins that can be combined with AI models in various ways, e.g., by using neural networks to represent the interrelationships of certain variables or by learning the optimal control of a complex system in a data-driven way. MoDigPro, for example, was a specific project. Here, we developed an AI for production control in an automotive plant and were able to achieve trouble-free continuous production as a result.

Where are the challenges?

Verena Wolf: So far, we have mainly worked with very simple ways of coupling different types of models. It is very difficult and research-intensive to design an optimal hybrid model for a given application since there are a variety of ways to combine mechanistic and neural models.
How do the application-oriented aspects of this research fit into DFKI's transfer ecosystem?

Antonio Krüger: Our transfer ecosystem of industry collaborations, Living Labs, TransferLabs, and the DFKI Academy training format bundles AI technologies at the interface of practical application. In the past, the question of how to solve a concrete problem, e.g., from production, was often answered with a recommendation for either a model-based or a data-driven method – depending on which had the fewest shortcomings for the respective application area. With Verena Wolf's approach of combining both model types, we have another hybrid approach in our portfolio that can be applied in the manufacturing industry or in the service sector.

Professor Wolf, you have developed a method that allows you to describe processes inside cells in much greater detail than was previously possible. In 2013, you were honored for this with Technology Review's Innovator Award. What is special about the approach you developed?

Verena Wolf: At the time, we developed an approach to describe the chemical reactions in cells at different levels of abstraction and to couple them accordingly. One can imagine this in such a way that particularly important areas of the system are viewed or modeled at a higher resolution, so to speak.

Why is this approach superior here, and what are you currently researching?

Verena Wolf: If you model all parts of the system in detail, the corresponding model becomes too complex. If one uses an abstract description, it becomes too imprecise. Combined models describe only certain parts of a system in a very detailed way and can correctly relate the coarse dynamics of other, more abstract subsystems. Currently, I am working on neuro-mechanistic models for drug development, among other things. This involves, for example, predicting the properties of specific molecules based on their structure. Such computer simulations can replace or at least significantly simplify time-consuming and cost-intensive steps in drug development. In order to bring these methods into the application, I would like to start corresponding DFKI projects very soon.

You also run the InfoLab computer science lab for school students at the UdS and have campaigned for computer science to become a compulsory subject at Saarland schools from grade 7. What skills should school leavers have?

Verena Wolf: Children need to understand the basic operating principles of the digital world. This will enable them to move more competently in this world and, in particular, to better understand its opportunities and risks. This includes, for example, knowing how information is stored and processed in a computer and how it can be securely encrypted or sent over data networks. Students should learn algorithmic thinking, i.e., how to solve problems using algorithms and how to implement their ideas using computer programs. An understanding of artificial intelligence should also be present by the end of middle school at the latest because AI systems are now encountered by young people everywhere in everyday life.

And how does DFKI support young people in computer science?

Antonio Krüger: DFKI has been involved in STEM mentoring programs for many years, supports the “MINT Zukunft schaffen” initiative, and provides a Saarland STEM ambassador in the form of company spokesman Reinhard Karger. We regularly open our doors and invite school classes for informational visits, focusing on the application perspective of the technologies on display and offering young people an insight into computer science-related fields of activity beyond programming. However, we also support the development of general AI competence with the free-of-charge further education offer “KI-Campus”.

What advice do you have for young people who want to pursue a career in computer science research?

Verena Wolf: Before enrolling in the bachelor's program, interested students should inform themselves about the requirements and the various computer science-related programs. In addition to the classic computer science bachelor's degree, Saarland University also offers more specialized programs, such as the Data Science & Artificial Intelligence program or Bio- or Media Informatics.

What do they need to bring with them?

Previous knowledge of computer science is not necessary. A good working attitude is much more important because studying computer science is very work-intensive, especially in the first semesters, and requires good self-organization. Students with a good understanding of mathematics often find the first semesters easier.

Pupils become students. What career opportunities does DFKI offer young graduates?

Antonio Krüger: The increasing digitization of industry has led to a significantly higher demand for managers with sound computer science backgrounds. Beyond a career as a software developer, DFKI paves the way to doctoral studies, to research departments and management floors in industry, or to spin-offs. Young colleagues with a degree in computer science or a computer science combination can work at DFKI from the very beginning on projects with clear objectives, often with an industrial connection. In addition, the DFKI is a partner in the Software Campus, a program that qualifies IT specialists for management tasks.

Ms. Wolf, do you have any special advice for girls and young women?

Verena Wolf: Girls often have a false image of computer science or of working as computer scientists. You shouldn't let yourself be influenced by prejudices based on a "nerd image." As a computer scientist, you can help shape the digital world. You can make hugely relevant contributions to society in numerous application areas and have a very well-paid and varied job that requires creativity and team spirit.

Ms. Wolf and Mr. Krüger, thank you very much for the interview!

Share this post:

Press contact:

Heike Leonhard, M.A.

Corporate Communications, DFKI

Prof. Dr. Antonio Krüger, CEO DFKI
Prof. Dr. Verena Wolf, Head of Neuro-mechanistic Modeling research department

German Research Center for Artificial Intelligence
Deutsches Forschungszentrum für Künstliche Intelligenz