BiGSEM doctoral candidates (Profile Management) are required to attend the BiGSEM Colloquium (Management) during the semester. Every doctoral student presents his/her own work or discusses relevant papers at least once a year. Interested people are invited to join. The sessions start at 2pm in room U3-140.
We will provide you with the titles and further information prior to the presentations.
"Innovative Matheuristic for Large-Scale Optimization: A Case Study from ROADEF 2022"
The ROADEF challenge 2022 involves RENAULT's supply chains, which span over 40 plants in 17 countries and involve 1500 suppliers. Every week, 6000 trucks deliver parts from suppliers to plants. The filling rate of these trucks is critical since the inbound transportation annual budget is several hundred million euros. The objective is to pack a set of items from suppliers into stacks and load the stacks onto trucks for delivery to the plants, with the goal of minimizing (a) the number of trucks used and (b) the inventory in the plants due to early deliveries. Items have a delivery time window ranging from 1 to 5 days. It is possible to deliver the items early (e.g., at their earliest arrival time) to better fill the trucks, but early deliveries generate inventory costs for the plants. Inventory costs are highly sensitive at Renault, as they represent several hundred million euros. Therefore, the best solution is to deliver items to the plants at the latest possible time while minimizing the number of trucks used. In terms of data volume, a large instance can contain up to 260,000 items and 5,000 planned trucks, spanning a horizon of 7 weeks.
In this paper, we introduce an innovative matheuristic designed for optimizing large-scale assignment and container loading problems. Our algorithm effectively showcases its capabilities by yielding new globally optimal solutions, which have the potential to generate significant cost savings for RENAULT's supply chain operations, potentially amounting to millions of euros. By tackling the intricate optimization challenges posed by the ROADEF Challenge 2022, our approach offers a promising avenue for elevating supply chain management, improving cost efficiency, and enhancing overall operational performance within a vast industrial context.
Peter Limbach and Christian Stummer
"Two editors' experiences on the dos and don’ts in academic publishing"
07.02.2024 - in U3-140 - it will starts at 3pm (instead of 2pm)
"Using Stochastic Programming for Surgery Scheduling under Uncertainty”
The scheduling of elective surgeries highly depends on the available capacities of beds in the postoperative intensive care ward. While these capacities, with their high demands on equipment and staff, are limited, the development of the occupancy of the intensive care units is subject to various factors of uncertainty, such as the individual length-of-stay of a patient or emergency arrivals. Despite this fluctuating demand, we aim to optimize the utilization of capacities in the intensive care unit after a surgery.
For this purpose, we present a model that proposes a scheduling of elective patients for their stay in the ward while considering the mentioned factors of uncertainty. We model this problem as a stochastic programming approach with a planning horizon of one week, in which we approximate the uncertain and decision-dependent evolution of the ICU occupation in scenarios generated by methods from Machine Learning and simulation techniques. The model aims to schedule as many patients as possible one week in advance while respecting the risk of exceeding critical occupancy levels as chance constraints and therefore avoiding outcomes that would be considered unacceptable in practice.
In cooperation with a hospital in Bielefeld, the model's architecture is based on real-world implications, using historic data from patients who were admitted to the intensive care unit. To demonstrate the applicability of the model, we created instances that represent real-world planning problems and typical patient characteristics. Our results show the potential for significant improvements in surgery scheduling by optimizing the utilization of the necessary capacities in postoperative care. Additionally, the conceptualization as a stochastic programming model with chance constraints allows the implementation of risk-oriented policies for the scheduling and admission of patients.
24.05.2023 (2-4 pm): BiGSEM Colloquium, U3-140
Speaker: Elias Schede will talk about "Selector: An ensemble for automated algorithm configuration"
Abstract: Solvers contain parameters that influence their performance and these must be set by the user to ensure high-quality solutions are generated, or optimal solutions are found quickly. Manually setting these parameters is tedious and error-prone, since search spaces may be large or even infinite. Existing approaches to automate the task of algorithm configuration (AC) make use of a single machine learning model that is trained on previous runtime data and used to create promising new configurations. We combine a variety of successful models from different configurators to an ensemble that proposes new configurations. To this end, each model in the ensemble suggests configurations and a hyper-configurable selection algorithm down-selects them to match the number of configurations to try to the amount of computational resources available. Using multiple models leads to a more diverse search, since each model has a unique belief about good regions of the search space, from which it can propose configurations.
Speaker: Frederik Tolkmitt will talk about "The role of uncertainty in innovation diffusion of radically new products: An agent-based simulation study"
Abstract: When consumers are uncertain whether they have sufficient (unambiguous) information regarding a new product, they might delay the adoption decision to a later point in time. This uncertainty effect can have a major impact on the market diffusion of an innovation. Most prior models that account for consumers’ belief updating in such a setting capture this effect by resorting to some form of Bayesian learning and, usually, assume that the distribution of all possible information with respect to the attributes of the new product is normal and that the pieces of information received by individual consumers are independent draws from this distribution. Consequently, consumers’ uncertainty regarding their beliefs decreases with each additional piece of information (e.g., after talking with a peer or being exposed to advertisements). This strong assumption is convenient as it makes models of opinion dynamics analytically tractable, and it works in many instances. However, when two diverging opinions are prevalent among consumers in a certain market and, thus, receiving additional information potentially increases uncertainty of individual consumers, a different approach is required. Radically new products, for which consumers cannot lean on previous experiences, constitute a prime example for such markets on which the distribution of beliefs can be bimodal. We propose a suitable approach that can deal with the latter setting and we demonstrate the value added of this novel approach (in contrast to the traditional approach) through computational simulation experiments based on an agent-based market model of innovation diffusion.
21.06.2023 (2-4 pm): BiGSEM Colloquium, U3-140
Presentation of the GOR Bachelor Thesis Award and a short overview of the winning work by Paulina Heine [Chair: Kevin Tierney]
Speaker: Corinna Hesse [Chair: Peter Limbach]
Title "Capital allocation" - Abstract
Speaker: Luisa Liedtke [Chair: Kai Bormann]
Title "Uncertainty in the context of corporate and daily purpose: A multilevel investigation" - Abstract
Speaker: Annika Schaefer [Chair: Kai Bormann]
Title "Crafting meaning out of contrasts: How illegitimate work tasks ignite job crafting and contribute to meaningful work" - Abstract
12.07.2023 (2-4 pm): BiGSEM Colloquium, U3-140
19th of October 2022
Speaker: Henning Witteborg
Title: Co-Design-Study of a Mobile Application for Cochlear Implantation Rehabilitation
Patients who undergo a cochlear Implantation face the need of an intensive aftercare to support their speech and hearing therapy. In order to support this process of training and improving the use of the implant, our group of researchers and practitioners are about to test an algorithm-based application that is supposed to offer patient-specific exercise program on a mobile device. For a first test with a small group of patients, we would like to add co-design-aspects to the development process. To make this process successful, we would like to discuss questions like the following:
• How close can patients can be included in the development process?
• Are joint workshops with all test patients better than individual sessions?
• How many iterations of feedbacks and adaptions are reasonable?
• What are the requirements that patients can really feel like stakeholders in the process?
30th of November 2022
11th of Januray 2023
15th of February 2023
20th of April 2022
Speaker: Felix Hagemann
Title: "Stochastic Shift Scheduling"
18th of May 2022
Speaker: Mohsen Nafar
Title: "Novel heuristics for compiling approximate decision diagrams for combinatorial optimizing"
Spearker: Felix Hagemann
Title: "Stochastic shift scheduling"
15th of June 2022
13th of July 2022