Hintergrundbild
Navigation
2018-08-28

Lecture from Dr. Meinolf Sellmann (GE Global Research)

Dr. Meinolf Sellmann from GE Global Research will be giving a lecture on Wednesday, September 5, 2018 at 14:00 in U3-140 entitled "Input-specific Algorithm Tuning"

Everyone is invited to see Dr. Meinolf Sellmann give the following lecture about his research.

Speaker: Dr. Meinolf Sellmann (GE Global Research)
Date: Wednesday, September 5, 2018
Time: 14:00
Place: U3-140

Title: Input-specific Algorithm Tuning
Abstract: We review the current state-of-the-art in instance-specific algorithm tuning. Most algorithms have parameters, whose setting can greatly influence algorithm performance. Rarely, one default setting alone works well across the board for a wide variety of algorithm inputs. Consequently, methods have been developed for setting algorithm parameters at runtime, when the input is known.

In this talk, we show the results of a thread of research that spans over the last ten years, and which has resulted in 20 gold medals at international competitions. In particular, we walk the audience through methods for finding robust default settings by means of genetic algorithms. We show how to exploit surrogate learning to improve tuning performance. And we introduce the most recent improvements in self-tuning algorithm portfolios with recourse. 

Aktuelles

Gemeinsame Tagung der AG Wirtschaftsinformatik (GOR) und der WK Operations Research (VHB)

Der Lehrstuhl für Decision and Operation Technologies hält am 24/25 Januar 2019 eine Tagung zum Thema "Data Science and Operations Research". Anmeldungen sind nun möglich.

Paper accepted in European Journal of Operational Research

2019-04-23

Our paper "A Branch and Bound Approach for Large Pre-marshalling Problems" has been accepted to the European Journal of Operational Research. The paper was coauthored with Prof. Shunji Tanaka, Consuelo Parreno-Torres, Prof. Ramon Alvarez-Valdes and Prof. Ruben Ruiz. We provide new lower bounds and a specialized branch-and-bound approach for solving the container pre-marshalling problem, an optimization problem found in certain types of container terminals. Our exact approach is extremely fast, providing solutions to even large scale problems in under an hour.