Will Data Science meet Optimization?



Combinatorial optimization has developed amazingly over the past decades, on certain problems beating Moore's law by factors of three or four. Techniques for solving combinatorial optimization problems vary in degrees of completeness. Impressive results have been obtained in complete as well as in heuristic techniques. While problems handled by combinatorial optimization techniques typically are described on a few pages, data must be available to allow extraction of benchmark sets of instances to be used when developing an approach. This makes us distinguish between explicit and implicit information. Automatic techniques have been developed to set implicit implicit information to use while developing algorithms. This effectively boils down to a data science technique assisting to improve, create or select appropriate algorithms. Data science can be used to approach problems merely based on implicit information. Explicit knowledge can strengthen these techniques, and this is often indispensible. Some techniques, some of which still under development, allow to make information in data explicit leading to other opportunities for optimisationtechniques. In this talk, some examples will be presented to illustrate this situation and to speculate about open possiblities at the interface of Daata Science and Optimization.

Registration to Krzysztof Postek, postek@ese.eur.nl is required for availability of lunch.