Presentation of the bachelor's thesis by Ms. Mona Lilith Bräunig:
"Surrogate Microservice for Uncertainty Quantification"
Advisor: Prof. Dr. Robert Schleich
The presentation will take place on Thursday, April 25, 2024 at 2;15 p.m. in Seminarroom 1/200 (INF 205, Mathematikon, first floor).
Presentation of the bachelor's thesis by Mr. Henrik Strootmann:
"Vergleich portabler SIMD-Programmiermodelle für C++"
Advisor: Prof. Dr. Peter Bastian
The presentation takes place at Friday, April 19, 2024, at 11 a.m. in Room 1.414 in INF 205 (Mathematikon, 1. Floor).
Presentation of the bachelor's thesis by Mr. Richter:
"Visualization of Magnetic Reconnection in Three Dimensional Plasma"
Advisor: Prof. Filip Sadlo
The presentation will be done hybrid on Monday, April 15, 2024, 2 pm, c.t. at Mathematikon, INF 205, Heidelberg, Seminar Room 10.
For those, who intend to join online, please use this link:https://heiconf.uni-heidelberg.de/kro-hx3-yug
Presenatation of the bachelor's thesis by Mr. Alexander Kosnac:
"Quantity-centric Summarization Techniques for Documents"
Advisor: Prof. Dr. Michael Gertz
The presentation will take place on Wednesday, April 3, 2024, at 1pm online via Webex. If you are interested in participating in the presentation, please send me a short message (gertz@informatik.uni-heidelberg.de).
Presentation of the bachelor's thesis by Mr. Kevin Klein:
"Mandelbulb-Acceleration on FPGAs"
Advisor: Prof. Dr. Dirk Koch
The presentation will take place on Thursday, March 21, at 3:00 p.m. at ZITI (INF368) in room 514 (5th floor).
Presentation of the bachelor's thesis by Mr. Nicolas Hellthaler:
"Footnote-Augmented Documents for Passage Retrieval"
Advisor: Prof. Dr. Michael Gertz
The presentation will take place on Friday, March 8, 2024, at 3pm online via Webex. If you are interested in participating in the presentation, please send me a short message (gertz@informatik.uni-heidelberg.de).
Presentation of the bachelors's thesis by Ms. Vera Helene Pittermann:
"Design and Test of a JTAG Configuration Interface for FABulous FPGAs"
Advisor: Prof. Dr. Dirk Koch
The presentation will take place on Friday, March 01 at 15:00 at ZITI (INF368) in room 514 (5th floor).
Presentation of the bachelor's thesis by Mr. Zins:
"Benutzerfreundliche Visualisierung von Qualitätsabweichungen | Konzeption und Entwicklung einer intuitiven Rückverfolgung von Analysewerten in Produktionsprozessen"
Advisor: Prof. Filip Sadlo
The presentation will take place in presence on Monday, January 29, 2024, 4 pm, c.t. at Mathematikon, INF 205, Heidelberg, Seminar Room 10.
Presentation of the bachelor's thesis by Mr. Martin Koloseus:
"A Simple and Efficient Algorithm for Minimum Cuts of Weighted Graphs"
Advisor: Prof. Dr. Felix Joos
The Presentation takes place on January 25, 2024 in Seminarroom 1 at 1 p.m.
Presentation of the bachelor's thesis by Ms. Olvia Stanossek:
"Vulnerabilities in Medical Data Transfer based on Blockchain Technologies"
Advisor: Prof. Dr. Vincent Heuveline
The Presentation takes place on Monday, December 18, 2023 at 3 p.m. in meeting room 1.414 in INF 205 (Mathematikon, 1. floor)
Presentation of the bachelor's thesis by Mr Edgar Brotzmann:
"Vorschlagen von Akzeptanzkriterien ähnlicher User Stories"
Advisor: Astrid Rohmann
The Presentation takes place on Friday, December 15, 2023 at 1:45 p.m. in Room 2/200, second floor of Mathematikon (INF 205).
Presentation of the bachelor's thesis by Mr. Miro Rashid:
"Grenzen der Prädikatenlogik erster Stufe: Gödelscher Unvollständigkeitssatz und Satz von Löwenheim-Skolem"
Advisor: Prof. Dr. Felix Joos
The Presentation takes place on December 4, 2023 in Seminarroom 2 at 1 p.m.
Presentation of the bachelor's thesis by Mrs. Vivian Kazakova:
"A Topic Modeling Framework for Biomedical Text Analysis"
Advisor: Prof. Dr. Michael Gertz
The presentation will take place on Tuesday, November 28, 2023 at 3:30pm online via Webex. If you are interested in participating in the presentation, please send me a short message (gertz@informatik.uni-heidelberg.de).
Presantation of the bachelor's thesis by Mr. Simon Körner:
"Making a Program Synthesizer Learn from Output Corrections"
Advisor: Prof. Dr. Artur Andrzejak
The Presentation takes place on Monday, November 20, 2023, at 3 p.m. online https://heiconf.uni-heidelberg.de/and-cxf-mxg
Presantation of the bachelor's thesis by Mr. Frederick Vandermoeten:
"PandaRec - A Recife-Based Recommendation System for the Pandas Library"
Advisor: Prof. Dr. Artur Andrzejak
The presentation takes place on Monday, November 20, 2023, at 2 p.m. online https://heiconf.uni-heidelberg.de/and-cxf-mxg
Presentation of the bachelor's thesis by Mr. Ingmar Lowack:
"Low– and No–Code Generation of Business Web Applications"
Advisor: Prof. Dr. Artur Andrzejak
The Presentation takes place on Monday, November 20, 2023, at 1 p.m. online https://heiconf.uni-heidelberg.de/and-cxf-mxg
Presentation of the bachelor's thesis by Ms. Anna Matveeva:
"Qualitative Comparison of Web Viewers for Presenting Interactive 3D Objects"
Advisor: Prof. Dr. Susanne Krömker
The Presentation takes place on Monday, November 13, 2023, at 4:15 p.m. at Mathematikon, INF 205, Heidelberg, Seminar Room 10.
Presentation of the bachelor's thesis by Mr. Felix Wörner:
"Engineering Buffered Algorithms for NeighbourCover"
Advisor: Prof. Dr. Christian Schulz
The presentation takes place on Tuesday, October 17, 2023, at 4 p.m. If you are interested in participating, please send a short message to christian.schulz@informatik.uni-heidelberg.de.
Presentation of the bachelor's thesis by Mrs. Daniela Fichiu:
"Chunking with Dynamic Sparse Distributed Memory, Hyperdimensional Computing and Transformer Attention"
Advisor: Prof. Dr. Artur Andrzejak
The presentation takes place on Wednesday, September 27, 2023 online via https://heiconf.uni-heidelberg.de/and-cxf-mxg.
Presentation of the bachelor's thesis by Mrs. Mara-Eliana Popescu:
"Induced Matchings in Graphs of Maximum Degree 4"
Advisor: Prof. Dr. Felix Joos
The presentation takes place on Friday, September 22, 2023, at 10 am in room 2/414 (Mathematikon).
Abstract: Ein induziertes Matching eines Graphen G ist eine Teilmenge M seiner Kanten, so dass beliebige zwei Knoten, die inzident zu verschiedenen Kanten von M sind, mindestens Abstand 2 in G haben. Das klassische Matchingproblem ist in polynomialer Zeit lösbar, aber es wurde bereits gezeigt, dass das Finden eines maximal induziertes Matchings in beliebigen Graphen ein NP-vollständiges Problem ist. Eine Möglichkeit, mit der Schwierigkeit des Problems umzugehen, besteht darin die Optimalitätsanforderung zu lockern und die Kardinalität des größten induzierten Matchings eines Graphen, die sogenannte induzierte Matchingzahl, zu approximieren.
Das Ziel dieser Arbeit ist, eine neue untere Schranke für die induzierte Matchingzahl für Graphen mit maximalem Grad 4 und vorgegebener minimaler Länge zu präsentieren. Unsere Ergebnisse ermöglichen zukünftige Verallgemeinerungen für ähnliche untere Schranken für die induzierte Matchingzahl bei Graphen beliebigen maximalen Grades und ohne kurze Zyklen.
Presentation of the bachelor's thesis by Mrs. Carola Schreier:
"Optimierung eines GPT-3-Chatbots für Anamnesegespräche: Die Rolle von Trainingsdatensatzgröße und -qualität"
Advisor: Prof. Dr. Jürgen Hesser
The presentation takes place online on Friday, September 8th, 2023, at 1 pm via https://heiconf.uni-heidelberg.de/hes-32w-u4v.
Presentation of the bachelor's thesis by Mr. Robin Hack:
"Automatische Bildsegmentierung mit Hilfe von Domain Adaptation"
Advisor: Prof. Dr. Jürgen Hesser
The presentation takes place online on Monday, September 4, 2023, at 9 am via https://heiconf.uni-heidelberg.de/hes-32w-u4v.
Abstract: Durch den technologischen Fortschritt spielt Bildsegmentierung eine immer wichtigere Rolle. Eines der bekanntesten Anwendungsgebiete ist die Entwicklung autonome Fahrzeuge. Doch nicht nur hier, sondern auch in der Medizin spielt Bildsegmentierung eine wichtige Rolle. So kann die Diagnose von Krankheiten und Krebs bei Menschen maßgeblich unterstützt werden [4]. Jedoch ist für die Segmentierung von Zellen ein Experte nötig, um diese durchzuführen. Zusätzlich ist sie auch noch sehr zeitaufwendig. Neuronale Netzwerke stellen eine Möglichkeit dar, dieses Problem zu lösen. Die Resultate von Algorithmen, die Bilder segmentieren, werden stetig besser [13]. Die Voraussetzung, um gute Ergebnisse zu erzielen, sind jedoch viele Trainingsdaten, also von Hand annotierte Bilder. In dieser Arbeit wird untersucht, ob es möglich ist, ein Netzwerk auf einem Datensatz zu trainieren und damit bisher ungesehene Datensätze in guter Qualität zu segmentieren. Dazu werden Netzwerke zum einen auf gelabelten HeLa Zellen und zum anderen auf gelabelten Stammzellen von Mäusen trainiert. Anschließend sollen dann histologische Bilder der Medizinischen Fakultät Heidelberg damit segmentiert werden. Dabei gilt, dass umso ähnlicher die Datensätze sind, mit denen das Netzwerk trainiert wurde und die Datensätze, die segmentiert werden sollen, umso bessere Ergebnisse sind zu erwarten. Aus diesem Grund werden die Bilder mit den HeLa Zellen und den Stammzellen von Mäusen vorher eingefärbt (in der Fachliteratur Staining genannt), um den histologischen Bildern, die segmentiert werden sollen, ähnlicher zu sehen. Diese Methode wird Unpaird Image Translation genannt und ein spezielles neuronales Netzwerk, ein Cycle GAN wird dafür verwendet. Dieses wird mit den HeLa Zellen und den histologischen Bildern trainiert, um anschließend das Staining der HeLa Zellen und den Stammzellen von Mäusen durchzuführen. Da die Stammzellen von Mäusen den histologischen Bildern ähnlicher sehen, werden hier bessere Ergebnisse erwartet. Da die histologischen Bilder keine Label haben, ist es nicht möglich eine Metrik zu berechnen, wie gut die Segmentierung funktioniert hat. Deshalb werden die Ergebnisse der Segmentierung mit ihrem Original verglichen, um das Resultat der Segmentierung zu sehen. Anschließend wird diskutiert, wie bessere Ergebnisse erzielt werden können
Presentation of the bachelor's thesis by Mr. David Pohl:
"Zero-Shot Word Sense Disambiguation using Word Embeddings"
Advisor: Prof. Dr. Michael Gertz
The presentation takes place online on Monday, Agust 28, 2023, at 1 p.m. (s.t) via Webex. If you are interested in participating, please send a short message to gertz@informatik.uni-heidelberg.de.
Presentation of the bachelor's thesis by Mr. Bastian Müller:
"An Adaptable Question Answering Framework with Source-Citations"
Advisor: Prof. Dr. Michael Gertz
The presentation takes place online on Friday, August 25th, 2023, at 1 p.m. (s.t.!) via Webex. If you're interested in participating in the presentation, please send a short message to gertz@informatik.uni-heidelberg.de.
Presentation of the bachelor's thesis by Mr. Adrian Becker:
"LeniaBreeder: Pattern Discovery in Continuous Cellular Automata with NEAT and a Human-in-the-Loop"
Advisor: Prof. Dr. Felix Joos
The presentation takes place on August 25th, 2023, at 10 a.m. in room 2/414 in Mathematikon.
Abstract: Lenia, ein Zellulärer Automat, ist ein komplexes System, das für seine Empfindlichkeit für kleine Veränderungen in seinem Anfangszustand bekannt ist, was die Suche nach interessanten Objekten erschwert. Vorherige Arbeiten zur Suche nach neuen und vielfältigen Objekten in Lenia nutzen maschinelles Lernen, um Objekte danach zu klassifizieren,was ein Computer als unterschiedliche Klasse von Objekten erkennt. In dieser Bachelorarbeit präsentieren wir einen neuen Ansatz, der Menschen in die Entdeckung neuer Objekte einbindet. Durch die aktive Einbindung von Nutzern in den iterativen Prozess der Suche nach neuen Objekten, können die Nutzer ihre Präferenzen und Interessen verwenden, um zu steuern welche Objekte erzeugt werden.
Presentation of the bachelor's thesis by Mr. Loris Wilwert:
"Engineering Sampling-based Streaming Graph Partitioning Algorithms"
Advisor: Prof. Dr. Christian Schulz
The presentation takes place online on August 8th, 2023, at 10am via HeiCONF. If you're interested in participating, please contact christian.schulz@informatik.uni-heidelberg.de.
Presentation of the bachelor's thesis by Mr. WOuter Florian Zeelen:
"Performace analysis of Cupy for computational physics"
Advisor: Prof. Dr. Robert Strzodka
The presentation takes place on Thursday, July 27, 2023, at 4.15 pm in room 524, INF 368 (ZITI, fifth floor).
Abstract: Numpy is often used in computational physics for vector and matrix math. It however may take too long to use when working with large datasets. Cupy presents itself as an alternative Numpy implementation that while using the same Numpy code may be faster for large datasets. This thesis examines the performance differences between Numpy and Cupy for computational physics. We take a look at differences for various array sizes between the two, what Cupy struggles with, and how to optimize code to run better on Cupy. We will see many different ways to minimize the use of loops while coding with Numpy/Cupy. At last, we conclude with for which tasks it is advantageous to use Cupy over Numpy.
Presentation of the bachelor's thesis by Mr. Marvin Hanf:
"Constructing a beta-Skeleton of an alpha-Shape of Point Sets"
Advisor: Prof. Filip Sadlo and Dr. Susanne Krömker
The presentation will take place in presence on Monday, July 17, 2023, 4 pm, c.t. at Mathematikon, INF 205, Heidelberg, Seminar Room 10.
Presentation of the bachelor's thesis by Mr. Till Rostalski:
"A Generic Patient Similarity Framework for Clinical Data Analysis"
The presentation takes place online on Friday, July 14, 2023, at 11 am via Webex. If you're interested in participating at the presentation, please contact gertz@informatik.uni-heidelberg.de.
Presentation of the bachelor's thesis by Mr. Tillmann Fehrenbach:
"DiCo & CoViz A differentiable programming engine for Deep Learning"
Advisor: Prof. Dr. Felix Joos
Presentation of the bacehlor's thesis by Mr Jonas Roos:
"Automated Extraction and Visualization of User Feedback"
Presentation of the bachelor's thesis by Mr. Sam Olinger:
"Comparative analysis of different Software tools to segment MRI images and run a tDCS simulation"
Abstract
The number of patients suffering from neurological diseases keeps growing. For drug-resistant
patients, a non-invasive brain stimulation like transcranial direct current stimulation (tDCS) can
be a great alternative. By simulating them, more knowledge of the procedure can be obtained
and the treatment can be adjusted for each person individually. Three freely available programs
are being analysed in this thesis as they come out of the box: SimNIBS, ROAST and Sim4Life
Lite. The simulation with the same head model is run on these tools with as little modifications
to their settings as possible. The runtime, the user interface and experience, and the result of
the simulation is compared between them. Since it was not possible to generate a final electric
field result with Sim4Life Lite, only the results from SimNIBS and ROAST could be compared.
Of those, the mesh from SimNIBS is better fitted to the tissues and smoother, and the tissues
in the interpolated voxel grid are well defined compared to the ones from ROAST. SimNIBS has
an average runtime of the simulation of about 2 minutes compared to a runtime of 14 minutes
in ROAST. ROAST’s advantage is the quicker segmentation and mesh generation, which is
13.5 times faster than the algorithm SimNIBS uses, as well as the fact that it is a MATLAB
library, which can be a benefit depending on the context of the project. Although, for that last
reason, ROAST can only be used as a function in MATLAB, while SimNIBS has a graphical
user interface on top of the command line access. The mean error between all the voxels is
19.2%, dropping to 10.25% within only the brain and to 6.68% when only considering the white
matter. This large difference is due to the different segmentation and meshing algorithms used
and must be considered when using the tools for patient-specific simulations.
The presentation takes place online on Tuesday, June 20th, 2023, at 10 am via https://heiconf.uni-heidelberg.de/hes-32w-u4v.
Presentation of the bachelor's thesis by Mr. Justin Sostmann:
"Evaluation of New Floation Point Formats"
Presentation of the bachelor's thesis by Sven Leschber:
„Hierarchical Discrete Markov Chains for Flow Visualization“
Advisor: Prof. Dr. Filip Sadlo und Dr. Susanne Krömker
The presentation will take place in presence on Monday, June 5, 2023, 4 pm, c.t. at Mathematikon, INF 205, Heidelberg, Seminar Room 10.
Presentation of the bachelor thesis by Mr. Florian David Schmäling:
"Analyzing the BrakTooth family of experimental attacks on specific Bluetooth chipsets"
Advisor: Prof. Dr. Vincent Heuveline
The presentation takes place on Monday, June 5, 2023, at 2 pm online via https://heiconf.uni-heidelberg.de/heu-f23-tn9.
Presentation of the master's thesis by Mr. Dennis Pfleger:
"Implementing an exploit as a Metasploit module and investigating the exploit ranking mechanism"
Presentation of the bachelor's thesis by Mr. Dominik Buchmann:
"Development of the Bokkel App to gamify reading and learning experiences"
Advisor: Prof. Dr. Jürgen Hesser
Abstract: The use of mobile devices and applications is becoming increasingly popular, especially among young children. The market for learning apps for children is also growing, with many companies vying for the attention of young users. This master’s thesis focuses on developing an app called Bokkel App that aims to blend traditional storytelling with augmented reality to create an interactive and educational valuable experience for young readers. The development of the Bokkel App takes place in an agile software development process. This methodology allows for flexibility in the development process as well as the ability to adapt to changing requirements and user feedback. The methodology includes elements of data driven-design, rapid prototyping and test-driven design. The development process is presented transparently and each step is described in detail and critically evaluated. An augmented reality module is used to create an immersive and interesting experience for young readers. In addition, mini games are included in the app to provide additional functionality and motivation for young users. The Bokkel App is developed in cooperation with Book-Play, a Heidelberg start-up company that specialised in the creation of children’s books, and contains their characters called Bokkel. The target group of the Bokkel App are children between the ages of 5 and 8, which corresponds to first time readers and aims to create an engaging and educational valuable experience for them, making reading more engaging and interesting by making books digitally explorable. The aim of this work is to provide a transparent overview of the development process of the Bokkel App, with a focus on the methodology used and the results achieved.
The presentation takes place on Wednesday, May 24, 2023, at 12 pm online via https://heiconf.uni-heidelberg.de/hes-32w-u4v.
Presentation of the bachelor's thesis by Mr. Yunus Sahin:
"Audiovisual Representation of Data through Glyphs and Tones"
Advisor: Prof. Dr. Filip Sadlo, Dr. Susanne Krömker
The presentation will take place in presence on Monday, May 15, 2023, 5 pm, c.t.at Mathematikon, INF 205, Heidelberg, Seminar Room 10.
Presentation of the bachelor's thesis by Mr. Paul Dietze:
"Formula Classification and Mathematical Token Embeddings"
Advisor: Prof. Dr. Michael Gertz
The presentation takes place on Wednesday, May 10, 2023, at 1 pm online via Webex. If you are interested in participating at the presentation, please write a short message to gertz@informatik.uni-heidelberg.de.
Presentation of the bachelor's thesis by Mr. Jonathan Bernhard:
"Optimized Bit-serial Operations on 64-bit ARM Processors using TVM Compilation and LIKWID Profiling"
Advisor: Prof. Dr. Holger Fröning
Abstract:
Presentation of the bachelor's thesis by Mrs. Filippa Piazolo:
"Adaptive Information Visualization"
Advisor: Prof. Dr. Filip Sadlo, Dr. Susanne Krömker
The presentation will take place in presence on Monday, April 24, 2023, 4 pm, c.t. at Mathematikon, INF 205, Heidelberg, seminar room 10.
Presentation of the bachelor's thesis by Mr. Fabian Regnery:
"Mushroom weather: Predicting the occurrence of Karposoma with the help of artificial intelligence"
Advisor: Prof. Dr. Jürgen Hesser
Abstract:
Everyone who collects mushrooms on a regular basis knows this situation: As soon as you arrive in the forest, you are either surprised by a yawning void or you stumble upon a mushroom at every corner.
The reason for this is the often faulty prediction of the occurrence of Karposoma, the fruiting bodies of mushrooms by the individual mushroom picker, who lacks simple, reliable methods.
This work shows a machine-learning based approach to predict the occurrence of Karposoma and a mushroom weather application I built based on a Python web service.
From the data of a survey of more than 350 mushroom excursions in autumn 2022 and the combination with local weather data I was able to train my model and integrate it into a well usable mushroom weather application including my own database.
As a result, a usable mushroom weather application exists for test purposes, the prediction of mushroom occurrence based on weather data can serve as orientation for mushroom pickers.
The presentation takes place on Friday, April 21, 2023, at 11 am via https://heiconf.uni-heidelberg.de/hes-32w-u4v.
Betreuer: Prof. Dr. Jürgen Hesser
Betreuer: Prof. Dr. Jürgen Hesser
Presentation of the bachelor's thesis by Mr. Manuel Trageser:
"Flow-based Traffic Classification using Deep Vision"
Advisor: Prof. Dr. Vincent Heuveline
The presentation takes place on Thursday, April 20th, 2023, at 4.15 pm via https://heiconf.uni-heidelberg.de/heu-f23-tn9.
Presentation of the bachelor's thesis by Mr. Felix Daas:
"Analyzing Schemes for Secure and Memorable Password Generation"
Presentation of the bachelor's thesis by Mr. Joshua Kraft:
"Automated Theft Detection in Retail Stores“
Advisor: Prof. Dr. Carsten Rother
The presentation takes place on Monday, April 17, 2023, at 2.30 pm in room B 128 in Mahtematikon B-part HCI.
Presentation of the bachelor's thesis by Mr. Akin Yilmaz:
"Konstruktion und Anwendung von Commitment-Verfahren"
Advisor: Prof. Dr. Wolfgang Merkle
Abstract:
Commitment-Verfahren sind ein zentrales Konzept der modernen Kryptographie. Im Vortrag wird zunächst die Konstruktion solcher Verfahren aus Einwegfunktion und Pseudozufallszahlengeneratoren behandelt, anschließend die Anwendung von Commitment-Verfahren auf Auktionen, Münzwürfe über das Telefon und Zero-Knowledge-Beweise.
The presentation takes place on Friday, April 14, 2023, at 2.30 pm in room 1 in INF 205 (Mathematikon, 2nd floor).
Presentation of the bachelor's thesis by Mr. Ahmad Mohammad:
"Sachlich und zeitlich getrennte Disaggregation von Kennzahlwerten in einer Planungslösung am Beispiel von SAP IBP"
Advisor: Prof. Dr. Jürgen Hesser
Abstract:
In der aktuellen Lösung von SAP Integrated Business Planning for supply chain (SAP IBP), Version 2208, erfolgt die Disaggregation von sachlich und zeitlich aggregierten Kennzahlwerten auf alle Kombinationen der untergeordneten Zeitperioden und sachlichen Attributen in einem Schritt. Wird zum Beispiel ein Wert für eine Produktgruppe, die 5 Produkte enthält, und eine Kalenderwoche auf die dazugehörigen Attribute verteilt, dann wird der Wert in einem Schritt auf die 5 Produkte * 7 Tage = 35 untergeordneten Knoten disaggregiert. Das ist zwar schneller als zunächst zeitlich und danach sachlich, oder umgekehrt, zu verteilen. Kunden wünschen sich aber für bestimmte Use Cases eine flexiblere Lösung. Ein Beispiel ist die Resteverteilung. In dieser Arbeit beschäftige ich mich damit, in welchen Fällen eine sachlich und zeitlich getrennte Disaggregation sinnvoll ist und welche Anforderungen an die Implementierung daraus simulieren. Mit einem Prototyp in Form eines SQL-Konsolen-Skripts simuliere ich eine sachlich und zeitlich getrennte Disaggregation. Dafür nutze ich unterschiedliche Methoden der Simulation Engine und entwickle zwei unterschiedliche Lösungsansätze, die ich abschließend bewerte. Das Ergebnis ist eine zeitlich und sachlich getrennte Disaggregation, die einen flexiblen Lösungsansatz für das untersuchte Szenario liefert.
The presentation takes place on Monday, April 3rd, 2023, at 3 pm via https://heiconf.uni-heidelberg.de/hes-32w-u4v.
Presentation of the bachelor's thesis by Mrs. Lucy Marmé:
"A Bootstrap Approach for Classifying Political Tweets into Policy Fields"
Presentation of the bachelor's thesis by Mr. Hossein Samieadel:
"Kernelization"
Advisor: Prof. Dr. Wolfgang Merkle
Abstract: Several standard examples for kernelizaton techniques in the field of fixed-parameter tractability are presented.
The presentation takes place on Monday, March 27, 2023 in room 1 in INF 205 (Mathematikon, second floor).
Presentation of the bachelor's thesis by Mr. Derin Theiventhiram:
"Simulation einer vaskulären Intervention mittels einer Game Engine"
Advisor: Prof. Dr. Jürgen Hesser
Abstract: This thesis deals with the development of a real-time medical computer simulation for performing a vascular intervention. The software includes a virtual operating room that is equipped with controllable medical devices and a patient. The simulation also includes the visualization of the X-ray projection and the interaction with a guidewire model developed by Robin Mader as part of his master’s thesis. The visualization is based on
the use of external computed tomography data and the "Lambert-Beer’s Attenuation Law" which describes the attenuation of X-rays when passing through matter, or in this case, the patient’s body. The existing guidewire model is based on the "Cosserat Rod theory" and "Position and Orientation Based Dynamics". The software provides an interface that allows any guidewire models into the system. For the simulation, the game
engine "Unity" is used which offers a user-friendly graphical editor as well as extensive libraries to implement the desired functions and to create an interactive 3D scene. In addition, Unity’s graphics pipeline offers to shift complex calculations to the GPU. The program can be run in the development environment on Windows, Linux and MacOS. The 3D graphic program "Blender" models the operation room and for the interaction
of the guidewire model with the blood vessel, the open-source software "Slicer 3D" is used which generates a 3D model with a triangular mesh structure from the CT data. The purpose of the software is to provide the ability to test a vascular intervention in a virtual environment before it is performed on a real patient. In this case, it is possible for the user of the simulation to navigate the C-arm-shaped X-ray machine and the X-ray
table, to observe the X-ray projection on a monitor and insert the guidewire into the blood vessel with the help of a control panel. It can be seen that the performance of the simulation depends on severalfactors such as the size of the computed tomography data, the parameters of the guide wire model, the hardware on which the simulation runs, etc.
The presentation takes place online on Monday, March 27, 2023, at 10 am via https://heiconf.uni-heidelberg.de/hes-32w-u4v.
Presentation of the bachelor's thesis by Mrs. Jing Fan:
"Assessing Factual Accuracy of Generated Text Using Semantic Role Labeling"
Advisor: Prof. Dr. Michael Gertz
The presentation takes place on Friday, March 24, 2023 at 1 pm online via Webex. If you are interested in participating, please send a short message to gertz@informatik.uni-heidelberg.de.
Presentation of the bachelor's thesis by Mr. Marko Lukosek:
"Polynomial Kernel of Total Dominating Sets"
Advisor: Prof. Dr. Felix Joos
The presentation takes place on Friday, March 24, 2023 at 10 am in room SR 2 (Mathematikon, INF 205).
Presentation of the bachelor's thesis by Mr. Jan Straub:
“Local Iterative Optimization for Graph Bundling“
Advisor: Prof. Dr. Filip Sadlo
The presentation takes place on Monday, March 13, 2023 at 3.15 pm in room 10 in INF 205 (Mathematikon, fifth floor).
Presentation of the bachelor's thesis by Mr. Christoph Hollizeck:
"Design of force myographic based control for soft wearable assistive technology"
Advisor: Prof. Dr. Lorenzo Masia
The presentation takes place on Tuesday, March 7, 2023 at 10 am in room 443 at INF 368, fifth floor and online via webex
Presentation of the bachelor's thesis by Mr. Daniel Schäffer:
"Higher-Order Streamsurfaces"
Advisor: Prof. Dr. Filip Sadlo
The presentation will take place on Monday, March 6, 2023, at 5 pm in room 10 of INF 205 (Mathematikon, fifth floor).
Presentation of the bachelor's thesis by Mrs. Theresa Häberle:
"Discrete Morse Theory and Conforming Complexes"
Advisor: Prof. Dr. Susanne Krömker
The presentation takes place on Monday, March 6, 2023, 4:15 pm in room 10, Mathematikon 5th floor and via online via https://heiconf.uni-heidelberg.de/kro-hx3-yug
Presentation of the bachelor thesis by Mr. Leonard Marks:
"Visualization of Integral Curves in Plane Fields"
Advisor: Prof. Dr. Filip Sadlo
The presentation will take place on Monday, February 27, 2023, at 4.15 pm at room 10 in INF 205 (Mathematikon fifth floor)
Presentation of the bachelor's thesis by Mr. Ahmad Fadl Allah:
"Automated Enterprise Software Testing Using Game Simulated Data"
Advisor: Prof. Dr. Jürgen Hesser
Abstract:
The modern supply chain is a complex network of suppliers, manufacturers, distributors, and retailers, all working together to bring goods and services to market. This complexity increases as global trade expands and customer demands change rapidly. To effectively handle these challenges, companies need to adopt a good digital supply chain management strategy. The COVID-19 pandemic has highlighted the need for companies to be able to adapt quickly to fast-changing supply chains.By leveraging technology to increase visibility into the supply chain and to automate key processes, organizations can improve their ability to anticipate and respond to change.
Integrated Business Planning system (IBP) is a cloud-based planning software for supply chain management that uses real-time information to help companies or organizations respond more quickly to market and business volatility. In the practice, IBP typically uses the data saved in an (ERP) Enterprise-Resource-planning where all resources, products, plants, bills of materials and the available transportation lanes between plants are centrally saved.To test IBP properly and separated from (ERP) and to cover various cases, we need a dataset of a global organization. Since this data is highly sensitive, we would rather simulate such a global organization via a detailed realistic logistics simulation game Widelands. In this work we aimed to extract logistical data from the Widelands simulation and then use this data as input to test the SAP-IBP. SAP-IBP will then create a logistical plan based on the simulated data we uploaded.
The last step of the work is to integrate the SAP-IBP created logistical plan back into the simulation and observe its impact on one of the players of the Widelands game.
The presentation takes place online on Monday, February 20, 2023 at 4pm at https://uni-heidelberg.webex.com/meet/juergen.hesser
Presentation of the bachelor's thesis by Mr. Ben Danglmayr
"Identifying the heart region of Medaka using deep convolution neural networks for object detection"
Advisor: Prof. Dr. Vincent Heuveline
Presentation of the bachelor's thesis of Mr. Samuel Melm
"Unifying Regular Expressions with a Domain Specific Language"
Advisor: Prof. Dr. Artur Andrzejak
The presentation will take place online on Monday, January 23, 2023, 6 pm, at
https://uni-heidelberg.webex.com/meet/artur.andrzejak
Presentation of the bachelor's thesis of Mr. Sheich Essa Samer
"Transformers Encoder Based Language Model for Pathological German Domain"
Advisor: Prof. Dr. Jürgen Hesser
The presentation will take place online on Monday, January 23, 2023, 4 pm, at
https://uni-heidelberg.webex.com/meet/juergen.hesser
Abstract:
In this work, we aimed to develop a Bert model that is capable of understanding pathological German texts. To do this, we trained the model mainly on pathological text. Our assumption was that training the Bert model only using pathological German data would lead to improved performance in pathological domain tasks. To this end, we first trained a tokenizer to create sub-word tokens that can encode text and present it to Bert for training. Bert creates contextual embeddings that can be processed in downstream tasks. To evaluate our model, we used two different text classification methods: single-label classification and multi-label classification. The reason for this choice is that pathological texts often contain descriptions of multiple organs or diseases. With the small dataset at our disposal, we tested whether single-label classification performed better than multi-label classification in the case of a small and unbalanced dataset. We compared our models with the German base version of Bert, which is based on the Google Bert model. Our results showed that our model performed well in both single and multilabel classification tasks. Single-label classification achieved better results than multi-label classification, and overall our models performed similarly to the German Bert model.
Presentation of the bachelor's thesis of Mr. Julian Bischoff:
"Learning a language by conversations with Amazon Alexa"
Advisor: Prof. Dr. Artur Andrzejak
The presentation will take place online at https://uni-heidelberg.webex.com/meet/artur.andrzejak on Monday, January 16, 2023, 6 pm.
Presentation of the bachelor's thesis by Mr. Johannes Pfeil:
"Processing and Analysis of Single Cell Whole-Genome Bisulfite Sequencing and Reduced Representation Bisulfite Sequencing Data from Isocitrate Dehydrogenase 2 mutated TF1 Erythoblasts"
Advisor: Prof. Dr. Moritz Gerstung
The presentation will take place on Monday, November 28, 2022 at 2pm in the seminar room in the Mathematikon B, 2. OG, Berliner Str. 41.
Presentation of the bachelor's thesis by Mr. Steven Kollortz:
" Development of pressure sensing insoles for gait event detection"
Advisor: Prof. Dr. Lorenzo Masia
The presentation takes place online on Tuesday, December 20, 2022, at 2pm via Zoom. In case you are interested in attending, please contact alexander.schubert@ziti.uni-heidelberg.de to get the Zoom link.
Presentation of the bachelor's thesis by Mr. Egor Svezhintsev:
"Inferring DSL-to-code translation rule sets from examples"
Advsior: Prof. Dr. Artur Andrzejak
The presentation will take place online on Wednesday, December 14, 2022, at 4:30pm at https://uni-heidelberg.webex.com/meet/artur.andrzejak statt.
Presentation of the Bachelor's Colloquiums by Mr. Gero Plettenberg:
"Experimental validation for kinematic control of a soft wearable exosuit"
Advisor: Prof. Dr. Lorenzo Masia
The presentation takes place on Thursday, December 1, 2022, at 3pm in room 443, 4. OG, ZITI INF 368. You can also participate in the presentation online via the link https://uni-heidelberg.webex.com/meet/lorenzo.masia.
Presentation of the bachelor's colloquium by Mr. Larry Harbrecht:
"Funktionsweise und Grenzen von DPI Umgehungssoftware"
Advsior: Prof. Dr. Vincent Heuveline
The presentation will take place on Monday, November 28, 2022, at 1pm online at https://heiconf.uni-heidelberg.de/heu-f23-tn9.
Presentation of the bachelor's thesis by Mr. Mika Mauro Rother:
"Performance Analyse eines Software Pakets mit C++ und Python"
Advisor: Prof. Dr. Guido Kanschat
The presentation takes place on Monday, November 28, 2022, at 11:15am in seminar room 11 (5th floor) in the Mathematikon.
ONLINE Presentation of the Bachelor's thesis by Mr. Patrick Steil
"Arc-Flags for Public Transit Routing"
Advisor: Prof. Dr. Christian Schulz
Abstract: Routenplanung in öffentlichen Verkehrsnetzen ist ein wichtiges Problem des öffentlichen Lebens. Auf landesweiten Netzen sind die Abfragezeiten (Query Times) der derzeitigen State-of-the-Art Algorithmen allerdings zu langsam. Die Arbeit entwickelt einen neuen Algorithmus, der Trip-Based Public Transit Routing mit einer angepassten Version der Arc-Flags Beschleunigungstechnik verbindet. Experimente auf realen Verkehrsnetzen zeigen das der neue Algorithmus deutlich bessere Abfragezeiten hat.
The presentation takes place on November 8 at 1pm via HeiCONF statt. If you are interested in attending, please contact christian.schulz@informatik.uni-heidelberg.de.
Online presentation of the Bachelor's Colloquiums by Mr. Stefan Iov:
"Reducing annotator disagreement on open-ended questions through dataset-specific instructions"
Advisor: PD Dr. Daniel Kondermann
Online at https://us06web.zoom.us/j/86376379101?pwd=ZU55eTEzQW1nVTkyNzdsY1d6OHVCUT09
Meeting-ID: 863 7637 9101
Kenncode: 004398
ONLINE presentation of the Bachelor's thesis by Mr. Manuel Hammer
"Security Analyses of Vulnerable Hosts: Concepts and Methods for Mitigating 'Log4Shell'"
Advisor: Prof. Dr. Vincent Heuveline
The presentation takes place online on Thursday, November 3, at 3pm via https://heiconf.uni-heidelberg.de/heu-f23-tn9
ONLINE-Presentation of the Bachelor's Colloquium by Mr. David Bogusch:
"Digitale Anamnese als Computerspiel"
Advisor: Prof. Dr. Jürgen Hesser
Online at https://uni-heidelberg.webex.com/meet/juergen.hesser
Presentation of the Master's thesis by Mr. Nicolas Bender:
"Integration of Dense Depth Predictions for Image Synthesis with Diffusion Models"
Advisor: Prof. Dr. Carsten Rother
The presentation will take place on Monday, April 22, 2024 at 5:00 p.m. in seminar room 3/128, Mathematikon building B, Berliner Str.
Presentation of the master's thesis by Mr. Stephan Lenert:
"Building Conversational Question Answering Systems"
Advisor: Prof. Dr. Michael Gertz
The presentation will take place on Tuesday, April 16, 2024 at 9:30am online via Webex. If you are interested in participating in the presentation, please send me a short message (gertz@informatik.uni-heidelberg.de).
Presentation of the master's thesis by Mrs. Maike Spankus:
"Ultrasound Registration for Needle Insertion Simulation Validation"
Advisor: Prof. Dr. Jürgen Hesser
The presentation takes place online at Thursday, April 11, 2024 at 10 a.m. via https://heiconf.uni-heidelberg.de/hes-32w-u4v.
Presentation of the master's thesis by Mr. Marc Bürg:
"Real-time Estimation of Stability Metrics from a Markerless Motion Capture System"
Advisor: Prof. Dr. Daniel Häufle
The presentation will take place on Wednesday, April 10, 2024, at 1:00 p.m., in room 514, ZITI (INF 368) 5th floor.
Presentation of the master's thesis by Mr. Jens Beyermann:
"Normalizing Flows for Image Registration"
Advisor: Prof. Dr. Karl Rohr
The presentation will take place online on Tuesday, April 9, 2024, at 2:00 p.m.
If you are interested in taking part, please send me a short message (k.rohr@uni-heidelberg.de).
Presentation of the master's thesis by Mr. Alexander Lepper:
"Extracting and Matching Concepts of User Stories and Acceptance Criteria"
Advisor: Prof. Dr. Barbara Paech
The presentation takes place online on Wednesday, March 27, 2024 at 3 p.m.
If you are interested in participating in the presentation, please send an email to paech@informatik.uni-heidelberg.de.
Presentation of the master's thesis by Mrs. Antonia Schlieder:
"Gaze-Based Recommendation of Visualization Exploration Paths"
Advisor: Prof. Dr. Filip Sadlo
The presentation will take place on Tuesday, March 12, 2024, at 2:15 p.m. in seminar 10, Mathematikon, 5th floor.
Presentation of the master's thesis by Mr. Florian Fallenbüchel:
„Investigating Domain Generalization using Invertible Neural Networks“
Advisor: Prof. Dr. Köthe
The Presentattion takes places on March 6, 2024 at 12:00 on MS Teems:
Presentation of the master's thesis by Mr. Johannes Maul:
"Reconstruction Of Surface Characteristics Based On Ultrasound-Computer-Tomography"
Advisor: Prof. Dr. Filip Sadlo
The presentation will take place on Monday, March 4, 2024 at 4:00 p.m. in seminar room 10 in Mathematikon (5th floor).
Presentation of the master's thesis by Mr. David Scheid:
"Image Generation for Application in the Field of Architecture"
Advisor: Prof. Dr. Carsten Rother
The presentation will take place on Monday, March 4, 2024 at 10:00 a.m., in Mathematikon building B, room 3/102.
Presentation of the master's thesis by Mrs. Angelina Basova:
"Table Extraction from PDF Documents"
Advisor: Prof. Dr. Michael Gertz
The presentation will take place on Wednesday, February 28, 2024 at
3pm online via Webex. If you are interested in participating in the
presentation, please send me a short message
(gertz@informatik.uni-heidelberg.de).
Presentation of the master's thesis of Gabriel Rashidi:
"Implementation of a Federated Self-Configuring Medical Object Detection Framework and Analysis of the Effects of Domain Shift on Generalization"
Advisor: Prof. Dr. Klaus Maier-Hein
The presentation will take place online on Tuesday, February 27, 2024 at 3:00 p.m.
Zoom Meeting
https://dkfz-de.zoom-x.de/j/4447543301
Meeting-ID: tel:4447543301
Presentation of the master's thesis by Mr. Johannes Bohlig:
"Biologically Inspired Ray Tracing"
Advisor: Prof. Dr. Filip Sadlo
The presentation will take place on Monday, February 19, 2024 at 4:15 p.m. in seminar room 11, Mathematikon, 5th floor.
Presentation of the master's thesis by Mr. Marco Kowalzik:
"Visualization of Overtwisted Disks in R3"
Advisor: Prof Dr. Filip Sadlo
The Presentation takes place on Monday, February 12, 2024 at 5:15 p.m. in the Seminarroom 10, INF 205 (Mathematikon, 5th floor).
Presentation of the master's thesis by Mr. Mikail Cayoglu:
"Electric Field Simulation for Phantom Comparison"
Advisor: Prof. Dr. Jürgen Hesser
The Presentation takes place online on Monday, February 12, 2:00 p.m. via https://heiconf.uni-heidelberg.de/hes-32w-u4v.
Presentation of the master's thesis by Mr. David Trajkovski:
"Passwortschemata zur Erstellung sicherer Passwörter allein durch Menschenhand"
Advisor: Prof Dr. Vincent Heuveline
The presentation will take place on Wednesday, February 7, 2024, at 4:00 p.m. in room 1/414, Mathematikon, 1st floor.
Presentation of the master's thesis by Mr. Christoph Rindchen:
"Automated Search for Efficient Transformer Architectures"
Advisor: Prof. Dr. Artur Andrzejak
The presentation will take place online on Thursday, February 1, 2024 at 10:00 a.m.
https://heiconf.uni-heidelberg.de/and-cxf-mxg
Presentation of the master's thesis by Mr. Benjamin Tuna:
"Automatically Relating Feedback and Existing Requirements"
Advisor: Michael Anders
Presentation of the Master's thesis by Mr. Lukas Klinger:
"Enhancement of self-configuring deep learning segmentation frameworks towards annotation-efficient learning"
Advisor: Prof. Dr. Klaus Maier-Hein
The presentation will take place on Friday, January 12, 2024 at 12:30 p.m. in the seminar room of the AIH Cluster (04.MC.100) in Berliner Straße 41 and will be held in German.
Presentation of the master's thesis by Mr. Luis Wettach:
"Medical Electronic Data Capture at Home - A Privacy Compliant Framework"
Advisor: Prof. Dr. Michael Gertz
The presentation will take place on Thursday, December 21, 2023 at 2:15pm online via Webex. If you are interested in participating in the
presentation, please send me a short message
(gertz@informatik.uni-heidelberg.de).
Presentation of the Master's thesis by Mr Thomas Möller:
"Engineering Weighted Connectivity Augmentation Problems"
Advisor: Prof. Dr. Christian Schulz
The presentation will take place online via HeiCONF on Wednesday December 20, 2023 at 10 am. Please register if you are interested (at christian.schulz@informatik.uni-heidelberg.de).
Presentation of the master's thesis by Mr. Viet Tran Ba:
"Hyperspectral Surgical Image Synthesis with Latent Diffusion Models"
Advisor: Prof. Dr. Lena Maier-Hein
Mr. Tran Ba will give his talk in person on Tuesday, December 19, 2023 at 11 am in the seminar room SR 4 in the Mathematikon (Im Neuenheimer Feld 205, 69120 Heidelberg).
Presentation of the master's thesis by Mr. Jason Pyanowski:
"Semantic Search with Contextualized Query Generation"
Advisor: Prof. Dr. Michael Gertz
The presentation will take place on Thursday, December 14, 2023 at 11am online via Webex. If you are interested in participating in the presentation, please send me a short message (gertz@informatik.uni-heidelberg.de).
Presentation of the master's thesis by Ms. Marie Christin Litziger:
"Edge Sparsification for Centrality Computation"
Advisor: Prof. Dr. Christian Schulz
The Presentation takes place online on December 13, 2023 via Heiconf. Please contact christian.schulz@informatik.uni-heidelberg.de, if you are interested.
Presentation of the master's thesis by Mr. Philipp Göldner:
"Information Retrieval using Sparse Embeddings"
Advisor: Prof. Dr. Michael Gertz
Presentation of the master's thesis by Mr. Robin Mader:
"Generating React code from machine-readable specifications"
Advisor: Prof. Dr. Artur Andrzejak
The presentation will take place online on Thursday, December 7, 2023 at 10:00 a.m. https://heiconf.uni-heidelberg.de/and-cxf-mxg
Presentation of the master's thesis by Ms. Lisa Kuhn:
"Scalability of Bayesian Neural Network Inference Methods for Real-World Tasks"
Advisor: Prof. Dr. Holger Fröning
The Presentation takes place on December 5, 2023, 5 to 6 p.m. in INF 368 Room 524 and online https://eu02web.zoom-x.de/j/61259854549?pwd=TGl0NjNoZ1ppSm9sdDJiaHBoQTFpZz09.
Presentation of the master's thesis by Mr. Simon Petrik:
"Human-in-the-Loop Optimization for Assisted Sit-to-Stand Trajectories"
Presentation of the master's thesis by Mr. Martin Drawitsch:
"Comparing Methods for Neuron Instance Segmentation"
Advisor: Prof. Dr. Fred Hamprecht
The Presentation will take place online via Zoom on Friday, November 24, 2023 at 11:00 a.m.
Accessdata for the Meeting:
Presentation of the master's thesis by Mr. Benjamin Sparks:
"Advanced Multidimensional Type Inference for Python"
Advisor: Prof. Dr. Artur Andrzejak
The presentation takes place online on Tuesday, November 21, 2023 at 9:30 a.m. https://heiconf.uni-heidelberg.de/and-cxf-mxg
Presentation of the master's thesis by Mr Oliver Wenz
"From Tabular Data to Python Code: A Low Code Approach to Data Manipulation"
Advsior: Prof. Dr. Artur Andrzejak
The presentation takes place online on Monday, November 20, 2023
Presentation of the master's thesis by Mr. Holger Wünsche:
"Evaluation of OpenMP as device backend in hipSYCL"
Advisor: Prof. Dr. Vincent Heuveline
The presentation will take place on Friday, November 3, 2023, at 2:00 p.m. in seminar room 1/414 (Mathematikon, 1st floor).
Presentation of the master's
thesis by Veton Kajtazi on the subject of
"Deep Learning based Tabular Dataset Synthesization for Data Privacy Protection"
Advisor: Prof. Dr. Vincent Heuveline
The presentation will take place on Wednesday, October 25, 2023 at 9:00 a.m. in seminar room 1.414 (Mathematikon A, 1st floor).
Presentation of the master's thesis by Mr. Lukas Bockstaller:
"Multi-Stage Software Aspect Classification"
Advisor: Prof. Dr. Michael Anders
The presentation takes place on Tuesday, October 17, 2023, at 2.15 p.m. in room 2/200 in Mathematikon.
Presentation of the master's thesis by Mr. Julian Freyberg with the title:
"Structural and Logical Document Layout Analysis using Graph Neural Networks"
Advisor: Prof. Dr. Michael Gertz
The presentation will take place on Monday, September 11th, 2023 at 11:00 a.m (s.t.!) online via Webex. If you are interested in participating in the presentation, please send a short message to gertz@informatik.uni-heidelberg.de.
Recent advancements in Large Language Models (LLMs) offer promising opportunities for real-world text-based Artificial Intelligence (AI) applications, but the complexity of integrating textual data from mixed formats like Portable Document Format (PDF) still presents challenges. Document Layout Analysis (DLA) is the task of extracting data from such Visually Rich Documents (VRDs). It is usually divided into Structural and Logical Document Layout Analysis. The structural analysis of a document aims to identify homogeneous regions within the document. In contrast, the logical analysis intends to assign
functional class labels like headline, text, or table to areas of the document.
As VRDs are diverse, various approaches exist to extract data from them. While the research community mainly focuses on image-based
documents, PDF documents are arguably highly important for real-world use cases. Most state-of-the-art (SOTA) approaches in the DLA research field leverage complex, multi-modal vision-language deep learning models that require pre-trained text or visual encoders and/or large amounts of training data and resources. Spiked by advances in the field of Graph Neural Networks (GNNs) in recent years, a spin-off
branch of research leveraging graph document models to create simpler but still effective DLA models has emerged.
In this thesis, we introduce a novel PDF layout annotation tool, PDFLAT, that helps to fill the gap in PDF datasets. Further, we present a novel graph-based DLA model PDFGLAM, capable of performing
structural and logical layout analysis of PDF documents. In contrast to other models, this model allows for end-to-end DLA while only requiring small amounts of training data and minimal hardware
resources by exploiting the inherent information of the PDF format. It transforms documents into a low-dimensional graph data structure and leverages GNNs for their analysis. Our evaluation - based on three different-domain datasets annotated using the novel PDFLAT tool - yields promising results for in-domain and cross-domain performance, training duration, and low-shot training, further substantiating the potential of graph-based models for DLA tasks.
Presentation of the master's thesis by Mrs. Marina Walther:
"A Universal Online Social Network Conversation Relation Model"
The presentation takes place online on Thursday, September 7th, 2023, at 10 a.m. via Webex. If you are interested in participating at the presentation, please send a short message to gertz@informatik.uni-heidelberg.de.
Presentation of the master's
thesis of Jessica Kächele on the subject of:
"Predicting Plasma Cell Infiltration in Multiple Myeloma: A Deep Learning based approach using MRI data"
Advisor: Prof. Dr. Klaus Maier-Hein
The presentation will take place online on Tuesday, September 5, 2023 at 9 a.m. https://us04web.zoom.us/j/72432823739?pwd=hgBo53hZmgrpfAvhb2vxaDM0lSXv5I.1.
Presentation of the master's thesis of Denitsa Ilieva on the subject of:
"Analysis of the contribution of clinical texts to early prediction of sepsis"
The presentation will take place online via HeiCONF on Monday, September 4, 2023 at 10:00 a.m.
If your are interested, please send a short Email to
riezler@cl.uni-heidelberg.de.
Abstract:
Sepsis is a life-threatening condition characterized by the body’s immune response to infection, leading to organ failure and mortality.
Early detection and prediction of sepsis are important for effective treatment and improved patient outcomes. This study focuses on early sepsis prediction using a combination of physiological values and clinical text data. The aim is to evaluate the performance of a Transformerbased model proposed by Wang et al. [39] applied to the MIMIC-III dataset. In our work, we conducted several ablation studies to
investigate how removing specific words affects the predictions of the model. We also focused on the effects of removing words with high average attention weights. These words were found to have a significant impact on model performance compared to other medical terms associated with sepsis diagnosis, such as "sepsis" and "infection". However, it should be noted that the model may encounter difficulties in
distinguishing between medical terms that are not exclusively specific to sepsis diagnosis but also associated with other diseases. This work emphasizes the need to consider specific criteria and further investigate the effect of different terms on sepsis prediction.
Presentation of the master's thesis by Mr. Dennis Scherbatschenko:
"Ableitung von Anforderungen aus Nutzerfeedback"
Presentation of the master's thesis by Mr. Klemens Gerber:
"Automatic Enrichment of Company Information in Knowledge Graphs"
Advisor: Prof. Dr. Michael Gertz
The presentation takes place online on Friday, August 25, 2023, at 10 a.m. (s.t.) via Webex. If you are interested in participating at the presentation, please send a short message to gertz@informatik.uni-heidelberg.de.
Presentation der Masterarbeit von Frau Sophia Matthis:
"Multi-Aspect Exploration of Plenary Protocols"
Advisor: Prof. Dr. Michael Gertz
The presentation takes place online on Wednesday, July 27th, 2023, at 10 am via Webex. If you're interested in participating at the presentation, please contact gertz@informatik.uni-heidelberg.de.
Presentation of the bachelor's thesis by Mr. Florian Kohrs:
"Subdivision-Based Graph Drawing"
The presentation will take place on Friday, July 28, 2023 at 11:00 a.m. in seminar room 10 (5th floor Mathematikon).
Presentation of the master's thesis by Sophia Heck:
"Approximation Algorithms for the Minimum (Sliding Window) Temporal Vertex Cover Problem"
Advisor: Prof. Dr. Christian Schulz
The presentation takes place online on July 27th, at 10 am via HeiCONF. If you are interested in participating, please contact christian.schulz@informatik.uni-heidelberg.de.
Abstract:
Modern real life networks are often highly dynamic. Temporal graphs represent these changes in the network configuration through discrete edge appearances on a fixed set of vertices. In a temporal graph all these changes are known in advance until a maximal timestep, the lifetime of the graph. The classical problem of Vertex Cover aims to find a set of vertices such that all edges are covered by one of their endpoints. This can be naturally extended in the time changing setting to the Temporal Vertex Cover (TVC) and Sliding Window Temporal Vertex Cover (∆-TVC). In the TVC every edge is covered once over the whole lifetime, while in the ∆-TVC every appearing edge is covered once in every window of ∆ consecutive timesteps. Both extensions are known to be NP-hard. In this thesis, known (approximation) algorithms for these extensions are implemented and experimentally evaluated. In particular, we consider two approximation algorithms and one non-polynomial exact algorithm. The approximations have approximation ratios bounded by the maximal degree d of any subgraph in at a certain timestep, leading to a ratio of d and d − 1. Moreover, two new approximation algorithms for the restricted case of always star temporal graphs are presented, leading to a 2∆ − 1 and a ∆ − 1 approximation ratio, where ∆ is the sliding window size. For the experiments a temporal graph generator for certain temporal graph classes and a framework for solving the (∆-)TVC are introduced. The experiments verify the stated runtime and approximation ratios of the known algorithm and even provide some improvements made through the implementations. Further, we compare the computation of the ∆-TVC on restricted inputs of always star temporal graphs through the known approximation algorithms with the here presented new ones. We show that the new approximations outperform the known d − 1-approximation in shorter runtime even in some cases where ∆ > d.
Presentation of the Master's Colloquium by Mr. Miguel Heidegger:
"Analysis and Optimization of Unsupervised Code-to-Code Translation"
Examiner: Prof. Dr. Artur Andrzejak
Link to online session: https://uni-heidelberg.webex.com/uni-heidelberg/j.php?MTID=m3ff7a76365a8972d43804d47df8ffb79
Presentation of the master's thesis by Colin Otchere:
"Visualizing crowdsourced user feedback in SMARTAGE"
Presentation of the master's thesis by Mrs. Therese von Wulffen:
"Cell Nuclei Segmentation with Sparse Shape Prior"
Abstract:
This thesis deals with the automated segmentation of cell nuclei in medical images, which plays a central role in medical research and diagnostics. A particular challenge is the segmentation of overlapping nuclei.
To address this challenge, we develop an algorithm based on a sparse template method. We use the prior knowledge that cell nuclei are approximately elliptical. The method involves image preprocessing to improve segmentation quality, followed by image partitioning into regions. For each region containing non-elliptical, i.e., likely overlapping nuclei, a corresponding template dictionary is created that contains shapes that resemble elliptical nuclei. The algorithm solves a sparse minimization problem and selects the most appropriate templates for the nuclei in the region. A large part of the work is devoted to the creation of the dictionary since appropriate shapes can only be selected if the dictionary contains them. The performance of the proposed algorithm is evaluated in detail using synthetic and real images. The results show that the algorithm is able to accurately segment a significant portion of the nuclei. Although the overall performance may be slightly lower than some existing some existing methods, the algorithm provides a solid alternative, especially for overlapping elliptical nuclei.
The presentation takes place online on Wednesday, July 5th 2023, at 3 pm via zoom:
https://eu02web.zoom.us/j/67791972077?pwd=RXlVT3loTXVMMjBGRC8zcTFvNzJvUT09
Meeting ID: 677 9197 2077 / Passcode: 766922
Presentation of the master's thesis by Mr. Pascal Memmesheimer:
"Adding Interpretability to an Anomaly-Based Method for Deep Packet Inspection in Intrusion Detection Systems"
Advisor: Prof. Dr. Vincent Heuveline
Abstract:
Due to digitalization, securing networks has become instrumental for any organization. Intrusion Detection and Prevention Systems aim to detect and pre-
vent malicious activity or policy violations. One kind of Intrusion Detection and Prevention System is Suricata. It is a signature-based system because it
uses pre-defined signatures to detect malicious packets. In contrast to this system, anomaly-based models like autoencoders or neural networks have become prominent because they do not rely on pre-defined signatures and generally are even able to detect zero-day attacks. In this thesis, we will present
how well Suricata is able to detect malware in contrast to strongly fragmented malware. Furthermore, we create four anomaly-based methods, one of them leveraging thegenerative abilities of a Wasserstein Generative Adversarial Network + Gradient Penalty to create anomalies to combat class imbalance artificially. We then compare their performances on the CSE-CIC-IDS2018 data set and our previously created fragmented malware data. Moreover, we make a detailed analysis using Shapley Additive Explanations to interpret predictions of the best-performing model. Lastly, we argue that Shapley Additive Explanations can help in the task of dimensionality reduction for classifiers.
The presentation will take place on Wednesday, July 5, 2023 at 10 a.m. in seminar room 1/200 (Mathematikon, 1st floor).
Presentation of the master's thesis by Raphael Kirchholtes:
„Compression of 4D Point Clouds using a Statistics-Based Octree Approach“
Advisor: Prof. Dr. Filip Sadlo, Dr. Susanne Krömker
Abstract
We contribute an approach for a content-sensitive compression method of point-cloud time-series data. The compression is done with help of a spatio-temporal octree struc- ture, that is stored in a PostgreSQL database, alongside the point data. The spatial resolution of the point-cloud data remains unchanged after the compression and can be fully restored from the database. We replace classical bounding-volume based octree building with statistical evaluations of every tree node’s contents to identify leaf nodes in the tree. The compression of time-series data is done with help of comparisons be- tween temporally consecutive tree nodes and their statistical evaluations. With this we are able to reduce spatio-temporal redundancies throughout the time-series at a com- pression rate of more than 60 % on test and real data. Test datasets were generated with the virtual laser scanning tool Helios++, whereas a selection of epochs from the sandy beach-dune dataset from Kijkduin were used as real data.
The presentation will take place in presence and hybrid on Monday, June 19, 2023, 4 pm, c.t. at Mathematikon, INF 205, Heidelberg, Seminar Room 10, and via https://heiconf.uni-heidelberg.de/kro-hx3-yug
Presentation of the Master's Colloquium by Mr. Julius Rominger
"Virtual Reality Adaptation for Bimanual Exoskeleton in Wearable Technology"
Examiner: Prof. Dr. Lorenzo Masia
The colloquium will take place in the conference room, 4th floor, INF 368
Presentation of the master's thesis of Mr. Maximilian Küchle:
"Supervised Feature Extraction for Cardiomyoctye Profiling"
Advisor: Prof. Dr. Christian Schnörr
The presentation will take place online on Friday, May 26, 2023, at 10:30 a.m. https://eu02web.zoom.us/j/63448179178?pwd=R2dZUGhLcTRIcExWaDdheFdoRTdpUT09
Presentation of the master's thesis by Mr. Dennis Pfleger:
"Implementing an exploit as a Metasploit module and investigating the exploit ranking mechanism"
Presentation of the master's thesis by Mr. Christopher Bruckner:
"Multi-Feature Clustering of Search Results"
Advisor: Prof. Dr. Michael Gertz
The presentation will take place online on Wednesday, May 24, 2023, at 2 pm via Webex. If you are interested in participating at the presentation, please send a short message to gertz@informatik.uni-heidelberg.de.
Presentation of the master's thesis by Mr. Fabian Wolf:
"Self-Supervised Learning for Medical Hyperspectral Image Segmentation"
Advisor: Prof. Dr. Lena Maier-Hein
The presentation will take place online/hybrid on Tuesday, May 23, 2023 at 11:00 a.m.
Location:
Besprechungsraum im Mathematikon, Bauteil B, 4. OG Berliner Straße 41
Zoom:
https://us02web.zoom.us/j/89636055915?pwd=ZU50WlZFTmNKRE42L05DR2Jhcm9MZz09
Meeting-ID: 896 3605 5915
Kenncode: 36283104
Presentation of the master's thesis by Mr. Jonas Reinwald:
"Implementing a security dashboard for the medical data platform provisioning toolkit Kaapana"
Advisor: Prof. Dr. Klaus Maier-Hein
The presentation takes place on Wednesday, May 10, 2023, at 4 pm via Zoom: https://us02web.zoom.us/j/81444456223 (Meeting-ID: 814 4445 6223). If you are interested in participating at the presentation, please write a short message to am248@stud.uni-heidelberg.de.
Presentation of the master's thesis by Mrs. Sophia Hammes:
"A Neural-Based Approach for Link Discovery in the Process Management Domain"
Presentation of the master's thesis by Mr. Alexander Haas:
"Towards Computational Histopathology Representations from MRI in Prostate Cancer"
Advisor: Prof. Dr. Klaus Maier-Hein
Abstract: Computational approaches provide the opportunity to infer cancer characteristics across imaging modalities. Inferring histopathological properties from non-invasive imaging can help to reduce the need for invasive diagnostic procedures. We therefore present and evaluate a generative approach to synthesize computational histopathology representations for a given MRI region in prostate cancer. By providing an image instead of a numerical score, we address the need for interpretability of the model output. To study the multi-modal problem, we first preprocess a public dataset with T2-weighted MRI and H&E stained whole-slide images. To achieve approximate spatial correspondence between the modalities, we register the histopathology image to the MRI image with a deep learning based registration approach (Shao et al., 2021). We address the resolution difference by introducing a 1-to-n relation, associating one MRI patch to multiple high-resolution histopathology patches. To generate histopathology representations with high image quality, we base our approach on an adversarial framework and implement PathologyGAN (Quiros et al., 2020), which we extend with an encoder to project the MRI image on its latent space. For a given MRI region, we evaluate if the model can generate a representative histopathology representations with morphological similarity to the spatially corresponding histopathology region. We show the model's capability of translating a separable input signal into representations with benign and malignant morphology in a toy experiment with binary input. Transferring this to real MRI input reveals instabilities of the model and points to further optimization towards computational histopathology representations from MRI.
Presentation of the master's thesis by Mr. Fabian Kneissl:
"Time-Dependent Graph Modeling of Twitter Conversations"
Advisor: Prof. Dr. Michael Gertz
The presentation takes place on Monday, April 3rd, 2023 at 2 pm online via Webex. If you are interested in participating at the presentation, please send a short message to gertz@informatik.uni-heidelberg.de.
Presentation of the master's thesis by Mr. Jannik Petersen:
"Planning and Implementing a Cloud Database with agile Methods for EDF scans"
Advisor: Prof. Dr. Jürgen Hesser
Abstract:
Electroencephalography (EEG) is a prominent method to study brain activity. The applications of EEG sensors extend from research studies to medical applications. In order to be able to search the data of the sensors categorized and sorted according to characteristics, software systems are used. Providing a central interface for storage also allows the data to be accessible for multiple medical studies. This master thesis describes the planning and development of a system to store EEG data in the European Data Format (EDF) in a database via a web-based user interface. For this purpose, a use case diagram is first sketched, which is then used as the basis for the architecture considerations. The application is implemented incrementally based on the use cases using the ASP.NET framework as a web development technology and agile methods. Apart from the EEG data storage, a syntactic analysis is implemented, which allows to read out individual data fields of the EEG data.
The presentation will take place on Thursday, March 9, 2023 at https://uni-heidelberg.webex.com/meet/juergen.hesser
Presentation of the master's thesis by Mr. David Hector
"High-Performance Randomized Eigensolvers"
Advisor: Prof. Dr. Peter Bastian
The presentation will take place on Thursday, February 9, 2023 at 11 a.m. in room 1.414, Mathematikon 1st floor.
Presentation of the master's thesis by Ms Fenja Schweder
"Probabilistic Modelling of Time Series for explorative Big Data Analysis in Astronomy"
Advsior: Prof. Dr. Vincent Heuveline
The presentation takes place online on Wednesday, February 1, 2023
https://heiconf.uni-heidelberg.de/heu-f23-tn9
Presentation of the master's thesis by Mr. Peter Hügel:
"Improving the Accuracy of Neural Networks when Predicting the Forward Propagation of Acoustic Waves"
Advisor: Prof. Dr. Jürgen Hesser
The presentation takes place on Thursday, January 26, 2023, at 4 pm in the seminar room 1/414 in INF 205 (Mathematikon, 1st floor).
Abstract:
The acoustic wave equation is an important part of state-of-the-art imaging methods. This
presentation explores the limitations of using deep learning to solve the forward propagation of
acoustic waves in heterogeneous media within the frequency domain. With the intention of
using the Trefftz method to solve the acoustic wavefield in a large domain, small subregions
are solved with a deep convolutional neural network. For this reason, the task is designed
so that the values at the boundaries of the patches are assumed to be known. With the goal
of predicting local Green’s functions within these smaller patches, networks are trained with
different configurations on a variety of datasets, taking into account the sound speed values
inside the patches. The accuracy is significantly improved by exploring different avenues.
Limiting factors are identified and possible future solutions to further reduce the error are
discussed.
Presentation of the Master's Colloquium by Mr. Fabio Gebhard
"A Rule-Based Approach for Numerical Question Generation"
Advisor: Prof. Dr. Michael Gertz
The presentation takes place on Monday, December 19, at 10 am (s.t.!), online via Webex. If you are interested in attending the presentation, please send a short message to gertz@informatik.uni-heidelberg.de.
Presentation of the Master's Colloquium by Mr. Max Goebels:
"Ergonomics of assistive exoglove control - Evaluation of sensory control interfaces for exogloves for patients with upper body motor disabilities"
Advisor: Prof. Dr. Lorenzo Masia
The presentation takes place on Wednesday, December 14, 2022, at 11am in room 443, 4th floor, INF 368.
Presentation of the master's thesis by Mr. Mike Söhner:
"Evaluation of hybrid parallelism in iterative linear solvers"
Advisor: Prof. Dr. Peter Bastian
The presentation takes place on Wednesday, December 14, 2022, at 10am in the seminar room 1/414 in INF 205 (Mathematikon, 1. OG).
Presentation of the master's thesis by Hamza Polat:
"Entwicklung einer Low-Code Webanwendung mit Hilfe von SAP-Programmiermodellen"
Advsior: Prof. Dr. Barbara Paech
The presentation takes place on Monday, December 12, 2022, at 11am in room 0/200 (ground floor) in the Mathematikon (INF 205).
Presentation of the master's thesis by Mr. Roman Erhardt:
"Engineering Algorithms for the Weighted Maximum Clique Problem"
Advisor: Prof. Dr. Christian Schulz
The presentation takes place on December 5, 2022, at 10am online via HeiCONF. In case you are interested in attending the presentation, please contact Christian Schulz at christian.schulz@informatik.uni-heidelberg.de.
Abstract:
The maximum weighted clique problem (MWC) is a well-known problem in graph theory with many applications. In this work, both exact and heuristic algorithms, which interleave successful techniques from related work with novel graph reduction rules are proposed. While graph reductions based on upper bounds have been used for MWC in the past, we present reduction rules, that make use of local graph structures in order to identify and remove vertices and edges without reducing the optimal solution. A set of exact reduction rules is employed in an exact algorithm called MWCRedu, while heuristic reduction techniques based on machine learning models such as MLP, Deepset and GNN are explored in the heuristic framework MWCPeel. Experiments on a broad range of graphs show, that MWCRedu outperforms the current state-of-the-art exact solver TSM-MWC [25] for most inputs. Specifically for naturally weighted, medium-sized street network graphs and random hyperbolic graphs, which are considered to model real-world graphs well, MWCRedu is faster by orders of magnitudes. The heuristic solver MWCPeel also outperforms its competitors FastWCLq [12] and SCCWalk4l [47] on these instances, but is slightly less effective on extremely dense or large instances. It is planned to publish the result.
Presentation of the master's colloquium by Mr. Armand Rousselot:
"Inverting Quantum Neural Networks"
Advisor: Prof. Dr. Ullrich Köthe
The pesentation takes place on Monday, November 12, 2022, at 2:30pm in the seminar room at the HCI, Berliner Straße 43 (Mathematikon B), 3. Floor and also online via MS Teams at https://teams.microsoft.com/l/meetup-join/19%3a208e6da633c1418bb04996827567e4fe%40thread.tacv2/1586168396078?context=%7b%22Tid%22%3a%223d057f2f-8466-43fd-8a56-5d334d13d895%22%2c%22Oid%22%3a%22315a4dcb-73da-4fc2-8eae-29ec654c41e0%22%7d.
Presentation of the Master's Colloquium by Mr. Yannick Pauler:
"State-of-the-art Self-supervised Monocular Depth Estimation for Outdoor Traffic Scenes: A Review"
Examiner: PD Dr. Bogdan Savchynskyy
Time & place: 2:30 p.m., Mathematikon, Building B, Berlinerstr. 43, Room B128
Presentation of the Master's thesis by Mr. Marius Bartsch
"Using Information Theory for Classification of Chatbot Requests"
Advisor: PD Dr. Wolfgang Merkle
The presentation will take place online on Monday, October 31, 2022, at 12:15 Uhr via https://heiconf.uni-heidelberg.de/7y93-ycxt-yp4y-zrzt
Presentation of the Master's Colloquium by Mr. Severin Laicher.
"Learning and Exploring Similarity of Sales Items and its Dependency on Sales Data"
Advisor: Prof. Dr. Michael Gertz
The presentation takes place on Friday, October 28, at 10 am (s.t.!), online via Webex. If you are interested in attending the presentation, please send a short message to gertz@informatik.uni-heidelberg.de.
Presentation of the Master's Colloquium by Mr. Erik Koynov:
"Weakly Supervised Cell Segmentation"
Examiner: PD Dr. Karl Rohr
If you are interested in participating, please send a mail to k.rohr@uni-heidelberg.de
Presentation of the Master's Colloquium by Giao Leo Nguyen:
"Machine Learning for Battery Design: Parameter Prediction withUncertainty Estimation for Partial Differential Equations“
Advisor: Prof. Dr. Ullrich Köthe
Time & place: 3:30 p.m., Seminar room HCI, Berliner Straße 43, Seminarraum B128
Presentation of the Master's Colloquium by Mrs. Veronika Nguyen:
"Unsupervised learning of a Graph-Multi-Matching Problem"
Advisor: PD Dr. Bogdan Savchynskyy
Time & Place: 14:30 Uhr, Mathematikon, Bauteil B, Berlinerstr. 43, Room B128
