Finding a solution with multiple conflicting goals is tricky. This is the process that Henrik Smedberg, a PhD student at the University of Skövde, has delved into in his research, which focuses on analyzing and extracting knowledge from all possible solutions. Even a poor solution can teach us something about the problems we are trying to solve.
Imagine you are faced with a tough decision involving several conflicting goals. You want to achieve the best outcome on all fronts, but compromises are inevitable. What is good for one goal is bad for another. This is where multi-objective optimization comes into play.
Let's say you are on the hunt for the perfect ice cream, but you have different preferences for taste, cost, and calories. With multi-objective optimization, you can generate a number of ice cream options that balance these preferences in different ways. Multi-objective optimization is a way to try to find the best possible solutions to problems when there are multiple criteria to consider and is commonly used in industries like manufacturing. Some powerful algorithms can solve such problems, but they generate many nonsensical solutions during the optimization process.
"My research focuses on analyzing and extracting knowledge from all solutions, both good and bad so that decision-makers can better understand the optimization process and learn from the results," says Henrik Smedberg.
Valuable insights are lost
Traditionally when dealing with multi-objective optimization, the sub-optimal, or bad, solutions are often discarded in the analysis of the results. This means that valuable insights can be lost in the process.
"In my research, I try to consider all solutions to gain knowledge that can help describe the characteristics of the problems. Even from bad solutions, we can find design principles that can directly describe optimal solutions, which would be very valuable for decision-makers to know," says Henrik Smedberg.
Mimer – a path to better decisions
An important result of Henrik Smedberg's research is the methods he has developed over the course of his work. They are general and can therefore be used for many different cases of multi-objective optimization. Furthermore, the research has resulted in Mimer, an openly available, interactive decision support platform that has already been used in the industry several times.
"I hope that my research can inspire practitioners in the field to focus more on knowledge discovery to better understand the solutions generated by the algorithms. An informed decision-maker can make more well-founded decisions," says Henrik Smedberg.
Henrik Smedberg will defend his thesis, "Knowledge discovery for interactive decision support and knowledge-driven optimization", at ASSAR industrial innovation arena, the University of Skövde on Wednesday, September 27, 2023.
Read more:
Mimer - Insights into the "black-box" optimization process
Try Mimer