Multiobjective Optimization Group

Faculty of Information Technology

University of Jyväskylä, Finland

The Multiobjective optimization group of the University of Jyväskylä, Finland, is a part of the Faculty of Information Technology and is headed by Prof. Kaisa Miettinen (since 1998). The research interests of the group are focused on (nonlinear) multiobjective optimization in the presence of conflicting objectives, including e.g.

• method development, with focus on
  • interactive methods
  • evolutionary and hybrid methods
• theoretical aspects
• software development
• real-world applications
  • simulation-based problems
  • data-driven problems (prescriptive analytics)

Overall, the work is inspired by real-life applications with the interest of supporting decision making. Interactive multiobjective optimization methods are at the very core of the interests of the group.

The Multiobjective Optimization Group is one of the few groups that specializes in implementing interactive methods, in particular, as open access software. A DESDEO framework is under preparation.

Besides simulation-based optimization, the group is interested in data-driven decision support and in particular prescriptive analytics whch we call decision analytics when multiple conflicting objectives are considered in making recommendations as decisions based on data available. We are active in the new thematic research field of JYU: Decision Analytics unitizing Causal Models and Multiobjective Optimization (DEMO) (with Kaisa Miettinen as the director). We are also interested in employing artificial intelligence and machine learning and, in particular, explainable artificial intelligence.

In the group, in the general theoretical and methodological development, the focus is typically on methods which are suitable and applicable in the case of industrial applications. Even though the methods applied are typically based on strong mathematical foundations, in practice, the applications may lack nice mathematical structures (they can be e.g. black box models and computationally expensive) and these practical characteristics must be taken into account when developing methods. Another characteristic of the methods developed is application-independence. In other words, behind the application-specific user-interface, the optimization method can be the same for designing paper machines or planning radiotherapy treatment.

Among others, the industrial applications considered deal with improvement of product properties, making production processes and their controls more efficient, or finding the best shape or structure etc.