02.12.2019: AdONE Seminar

Prof. Stefan Helber (Leibniz-Universität Hannover): "Pros and cons of service level metrics for stochastic lot sizing models" and Anja Kirschbaum (TUM, AdONE): "Learning Convex Polytopes"


Prof. Stefan Helber (Leibniz-Universität Hannover): "Pros and cons of service level metrics for stochastic lot sizing models"

Lot sizing decisions in a manufacturing context are often made for multiple products over multiple discrete periods. Very often, the different products compete for scarce production resources. When those decisions are made, the demand for the products is often uncertain, e.g., as it stems from the application of a forecasting model. If limited production resources are confronted with uncertain product demands, it can be impossible or economically unattractive to meet the entire demand in time. Back-ordering can be employed such that customers have to wait until their demands are actually fulfilled. Different types of service level metrics have been proposed to quantify the service provided by the production system in such a situation. Those metrics can be embedded into discrete-time, multi-product lot sizing models to develop production schedules that are economically attractive, feasible with respect to the capacity constraints, and ensure a certain service level in spite of this stochastic demand. In this presentation, we will discuss different service level metrics in the context of the stochastic capacitated lot sizing problem (SCLSP), in particular those that are related to expected backlog and waiting times. We will furthermore show how to implement them in optimization models using either scenario techniques or linearization techniques of non-linear functions of expected values of backlog and physical inventory.

 

Anja Kirschbaum (TUM, AdONE): "Learning Convex Polytopes"

Data transformations are a standard tool in machine learning. We consider a novel type of data transformation, which has already shown its potential in practical studies. In this talk, we first define and motivate the data transformation in a general setting, but the main focus lies on its application to convex polytopes: Besides some basic properties of the transformation, we discuss several (classification) techniques which, combined with the transformation, can be used to learn convex polytopes.

 

Date: 02.12.2019, starting at 1:30 p.m.

Location: City campus 0505.EG.544