Achtung:

Sie haben Javascript deaktiviert!
Sie haben versucht eine Funktion zu nutzen, die nur mit Javascript möglich ist. Um sämtliche Funktionalitäten unserer Internetseite zu nutzen, aktivieren Sie bitte Javascript in Ihrem Browser.

Masters Thesis Topics

These are our currently available topics, although you are welcome to suggest your own ideas as well. All topics may be written in English or German.

Optimierung des Radar-Managements in Multi-Aircraft Missionen


Das Fraunhofer Institut für Hochfrequenzphysik und Radartechnik (FHR) in Wachtberg bei Bonn untersucht Konzepte, Methoden und Systeme für elektromagnetische Sensoren, insbesondere im Bereich der Radartechnik. Die FHR-Abteilung Kognitives Radar (KR) befasst sich mit Forschungsthemen an der Schnittstelle zwischen Informationsverarbeitung und Radartechnologie. Airbus Defence and Space (DS) ist ein Geschäftsbereich der Airbus Group, spezialisiert auf militärische Luftfahrt, militärische und zivile Raumfahrtsysteme sowie Sensoren und Kommunikationstechnologie für Verteidigung und Sicherheit. Die Airbus DS Abteilung „Future Projects“ befasst sich mit innovativen Flugführungs- und Sensor- managementkonzepten für zukünftige Airbus DS Produktgenerationen.

Im Rahmen einer Multi-Aircraft Simulation sollen mehrere Szenarien und Situationen durchgespielt werden, die verschiedene Anforderungen an das Radar-Management der Missionsteilnehmer stellen. Dazu wird bei Airbus DS derzeit ein Automated Planning Ansatz evaluiert. Zusätzlich soll im Rahmen der hier ausgeschriebenen Arbeit die genannte Funktionalität mittels Constraint Programming realisiert werden. Hierbei soll der Fokus neben der Multi-Aircraft Missionsplanung vor allem auf die Optimierung des Radar- Managements gelegt werden. Das Know-How hierzu wird vom Fraunhofer FHR bereitgestellt.

In diesem Kontext bietet das Fraunhofer FHR gemeinsam mit Airbus DS eine Bachelor- bzw. Masterarbeit an.

Aufgaben:

  • Entwicklung von Funktionalität für intelligentes, optimiertes Radar-Management in militärischen Missionen
  • Entwicklung von Entscheidungslogik für automatisiertes Verhalten von kooperativen Missionsteilnehmern


Ihre Voraussetzungen:

  • Interesse an Künstlicher Intelligenz, rationalen Agenten, Optimierungsproblemen und/oder moderner Entwicklungsmethoden
  • Interesse an Sensormanagement, Radartechnologie und Signalverarbeitung
  • Idealer Weise Kenntnisse in Constraint Programming, Logic Programming oder Mixed Integer Linear Programming (z.B. CPLEX, PROLOG)

Dieser Arbeit wird durch der Lehrstuhl für Entscheidungsunterstützungssysteme und OR betreuet. Weitere Fragen zur Thematik bitte an:

Herrn Dr. Stefan Brüggenwirth (FHR) stefan (dot) brueggenwirth (at) fhr.fraunhofer.de
Tel.: +49 (0) 228 9435-173

Herrn Ruben Strenzke (Airbus DS) ruben (dot) strenzke (at) airbus.com
Tel.: +49 (0)8459 81 67604

Algorithm configuration for the stock market

The company Quantopian has introduced a radical new strategy for managing a hedge funds. Instead of an expert manager leading a hedge fund and deciding what to buy and sell, Quantopian crowd sources its decisions, allowing anyone to write an algorithm to run their own hedge fund. In the case of Quantopian, selected algorithms are actively used on the stock market and a portion of the gains are paid to the creators of the algorithm.

The idea behind this thesis is to use an algorithm configurator to tune market trading strategies that can be used with the data freely provided in Quantopian. We will learn what strategies work well on the market and submit them to Quantopian. Furthermore, we will investigate “niche” strategies focused on specific portions of the market to raise the chances of the algorithm being included in Quantopian’s portfolio.

Programming knowledge is required, knowledge of the stock market is not. This topic can be used for a bachelors or masters thesis.

Analyzing differences between linear mixed-integer optimization models

Linear optimization models describe a set of feasible solutions from a set of inequalities called constraints. Given two models, identifying what is different between those models is a difficult task. For example, an interesting property of these models is that the order of variables and constraints does not change the set of feasible solutions. Therefore, permutations of the variables and constraints result in equal representations of the same model. Furthermore, different components of a linear model can change: variables and constraints (problem matrix), variable and constraint bounds as well as the objective function(s).

The goal of this work is to develop and implement a tool to compare different model representations and to identify and display differences (e.g. added constraints, deleted variables, variable type changes).

Tasks:

- Literature research about suitable methods for determining the equivalence between different representations of the same linear mixed-integer problem.

- Implementation of methods to detect and display a defined set of changes, e.g. 

  • change of objective function
  • added constraints
  • added variables
  • variable type change
  • variable bound change
  • constraint bound change
  • coefficient change of a constraint

- Validation of the tool on a set of test instances

Model reading is done using Gurobi Optimizer and we will provide demo models for validation.

Skills: Advanced knowledge of linear optimization,  analytical thinking, programming

Combining algorithm selection and configuration

Algorithm configuration techniques are used to find high quality parameter settings for parameterized algorithms and solvers. These parameter settings allow solvers to find good solutions faster, or find better solutions in a fixed amount of time.
Algorithm configuration has traditionally been applied to a dataset in one of two ways. The first way involves simply running the configurator on the entire dataset and seeing what parameters result. The second way partitions the dataset into multiple pieces and configures each piece separately, resulting in a portfolio of parameterizations. Although the first method is easy to apply, finding better parameters for some instances in the dataset could provide better results. The second option solves this issue, but methods doing this have very long runtimes.
We propose to make a new type of algorithm configurator that creates a portfolio for a dataset over a single execution of the solver. In this way, high quality, instance-specific parameters can be found for a dataset  without the expensive computation necessary for iterative or partition techniques.

Reproducibility in Operations Research

Reproducibility in scientific work has become a major theme in a number of areas of science. However, there has not been any close examination of how reproducible results are in Operations Research. Can the performance claims of articles in the field be confirmed? What is the quality of the experimental evaluations performed in the literature?

This thesis will investigate a number of aspects of reproducibility, starting with analyzing whether code and datasets from well-known optimization problems (such as the vehicle routing problem) are even freely available, or are made available upon request.    After collecting data and solvers, various experiments will be conducted to see if the results match what was previously reported.

We will use algorithm configuration and data mining techniques to view well-known approaches in a new light. Furthermore, we will use data mining techniques to assist in assessing and visualizing the performance of chosen algorithms.

Propose your own topic
AdvisorKevin Tierney
TopicIf you have an idea for a master's thesis, feel free to propose the idea to me. I am interested in all areas of decision support and operational research, and would be happy to consider your proposal for a thesis.
LanguageEnglish, German
TypeMaster Thesis
StatusAvailable

The University for the Information Society