Vortrag von Serpil Sayin

am 15.05.25 
um 14:00 
in 31-302 

wird Serpil Sayin einen Vortrag über Exact and Heuristic Representation Algorithms for Multiobjective Optimization Problems halten. Interessierte Gäste sind herzlich willkommen.

Abstract:
In most realistic settings, obtaining the complete nondominated set of a multiobjective optimization problem is computationally challenging. Moreover, presenting this set to a decision maker to identify a most preferred solution is often impractical. Representation algorithms address this challenge by seeking a representative subset of the nondominated set rather than the full set. While finding an arbitrary representation is typically straightforward, the quality of such representations is not always guaranteed.  Heuristic representation algorithms offer a balanced approach, providing reasonably good representations with modest computational effort. In contrast, exact representation algorithms aim to deliver solution sets that are guaranteed to meet a specified quality threshold, as defined by a given measure, even in the absence of the full nondominated set.
In this talk, we present an overview of an exact representation algorithm for discrete problems. This algorithm is adapted from a previous exact method that computes the entire nondominated set and is designed to operate with a coverage error guarantee. We highlight the challenges involved in ensuring a specified level of coverage error, describe our proposed solution, and summarize computational findings. We then transition to heuristic representation approaches and report on our current work, which focuses on reducing the number of objective functions in a given problem. We explore alternative strategies for performing such reductions and assess the nondominated set of the  lower-dimensional problem as a representation of the original nondominated set using several measures. We conclude by outlining potential directions for future research.