Jeremias Berg

Optimization problems are ubiquitous in modern academia and industry. Whenever we are faced with a problem where we are tasked with finding a “best” solution from a discrete set of feasible alternatives, we are solving a combinatorial optimization problem. Most of the interesting optimization problems are computationally challenging, i.e. NP-hard. Such problems arise in various different domains, including planning, scheduling, data-analysis and machine learning. As such, effective solution algorithms can save time, money or other resources in various applications.

My research focuses on developing general and effective solution methods to NP-hard combinatorial optimisation problems. I focus especially on so called declarative approaches for solving optimisation problem. In a declarative approach, the optimisation problem being solved is first modelled as a set of mathematical constraints. Afterwards a constraint algorithm (a solver) is used to find an optimal solution to the constraint instance, from which an optimal solution to the original problem is then constructed. In addition to developing constraint solving technology I also apply the created methods toward applying constraint-based methods for solving various real-world optimisation problems. Lately I have been especially interested in applying constraint-based methods toward explainable AI and machine learning, thus increasing the potential application domains of machine learning even further.

At the moment I am working as a postdoctoral researcher at the department of computer science of the University of Helsinki.
Home page:

Picture: Juuso Koivisto, Bonafide Creatives.