- This event has passed.
ALOP Colloquium with Kendra Reiter
August 4 / 16:00 - 17:00
On Thursday, August 4 2022 at 16:00 c.t. Kendra Reiter, MTU Maintenance (Hannover), will speak at the ALOP Colloquium about her recent work:
Building a geo-referenced microsimulation model with discrete optimization*
*joint work with Dr. Ulf Friedrich, OVGU Magdeburg and Prof. Dr. Ralf Münnich, Trier University
Microsimulation is an important tool to support evidence-based policies. To produce a fully geo-coded dataset, where information is generally available on aggregate levels of different hierarchies, micro units (households and persons) have to be placed into geo-coded dwellings. The microsimulation model involves mathematical optimization problems with integrality constraints on some of the optimization variables. It is therefore necessary to employ combinatorial optimization techniques to handle these discrete structures efficiently. More specifically, fast algorithms for the sub-problem of address selection are needed: Given a population generated in the first step of the microsimulation process and a target region, the households in the population have to be assigned to actual addresses within the region, i.e., an address has to be selected for each household in the population.
While the computation time is often not crucial when considering only a subset of the population, e.g., for the simulation of a certain region or city, the big-data setting of a complete model typically requires specialized, fast algorithms and techniques from data science. For example, in the address selection model for Germany more than 40 million households are assigned to over 25 million addresses while using several statistical variables to measure the quality of the assignment. General purpose heuristics such as simulated annealing do generally not solve this instance within an acceptable time limit and do not provide quality certificates. In addition, large data sets from several sources (e.g., Open Street Map, city registers, grid-based census data) have to be combined and pre-processed in an efficient and secure way.
Please join us at 16:00 c.t. in HS 9.