Few of us are aware of it on a daily basis, but we are in fact surrounded by so-called cyber-physical systems. Briefly explained, these systems comprise software and hardware components communicating with and controlling a physical reality. They include wind turbines, cars and objects as critical as pacemakers.
In these systems, the use of machine learning is very widespread – in other words, the systems are controlled by models learning from historical data.
However, the use of machine learning also leads to challenges that may have fatal consequences. Professor Kim Guldstrand Larsen from the Department of Computer Science at Aalborg University has just received a DKK 30 million Villum Investigator grant for a project aiming to ensure that cyber-physical systems meet requirements concerning reliability and safety to a much higher degree than is the case today. He explains:
- Machine learning is great for a wide range of systems. The problem arises when the system is subjected to a situation that has not been used in the training data. For instance, a Tesla may smash into the side of the road if it does not recognize the objects in its surroundings. The use of machine learning may lead to inexplicable situations that might have been identified up front, and which clearly violate obvious safety requirements.
Therefore, the aim of the project is to combine machine learning with mathematically sound and comprehensible techniques capable of offering absolute guarantees. In a manner of speaking, the researchers will set boundaries for what is possible when using machine learning:
- We provide it with a kind of safety shield that will ensure that we do not end up in situations where things go wrong. It is also a matter of being able to look much further ahead in the process. I am not saying that things will be 100 percent accurate, but I hope and believe that we will get a long way towards that goal.
DELVING INTO THE THEORETICAL BASIS
First of all, Kim Guldstrand Larsen is looking forward to kicking off the project, which will have a strong focus on models, in collaboration with colleagues from the research group DEIS – Distributed, Embedded and Intelligent Systems:
- We will be doing a lot of work on models and interactions between components that are relevant in a physical reality. It will of course be a crucial point for the entire project that we transform these models to algorithms, and later to tools that engineers working in different societal contexts can use.
In addition, Kim Guldstrand Larsen sees great advantages in the fact that the group is already doing significant work within fields where cyber-physical systems are used – particularly within energy and transport.
- We will build on the fact that we have a wide range of projects in which we can test things in practice. But for the moment, I am happy that we now have plenty of time and opportunity for delving into the theoretical basis and building this shield. I hope that by the end of the project, our research group will be even more excellent, with several colleagues having contributed to creating entirely new techniques that I do not even know today. Techniques and tools that will be used by us, by colleagues within other fields and, of course, out in the real world.
ABOUT VILLUM INVESTIGATOR
This year, the Villum Investigator programme awards DKK 313 million to ten experienced and internationally renowned researchers who will continue to spearhead world-leading research. The researchers have been selected as Investigators among 72 applicants – following the international practice of assessment of the applications by the VILLUM FONDEN’s working group, assessment by three independent peers and final selection by the foundation’s board.
On 27 April 2021, the Villum Investigators were celebrated at a ceremony in VILLUM Window Collection in Søborg with participation of, among others, the Danish Minister for Higher Education and Science, Ane Halsboe-Jørgensen.
The project S4OS: Scalable analysis and Synthesis of Safe, Small, Secure and Optimal Strategies for Cyber-Physical Systems will have a duration of six years, starting from September 2021.
Phone: +45 2090 1829 /+45 2294 0459