Themenliste
Diese Liste beinhaltet Themen für potentielle Bachelor- oder Masterarbeiten, aber auch Projekt-, HiWi- oder andere Tätigkeiten. Suchen Sie mit Str+F nach gewünschten Keywords, wie z.B. 'machine learning' oder ähnliches.
Selbstverständlich sind auch eigene Ideen jederzeit willkommen! Wenden Sie sich dafür bitte mit kurzem Lebenslauf, Notenspiegel und evtl. Themenvorschlägen an Laura Maria Schaal. Wir melden uns in Kürze bei Ihnen!
Thema:
Analysis and Visualization of Urban Spatial Data
Art der Arbeit:
Master, Project, HiWi, Bachelor
Betreuer:
Beschreibung:
Spatial units such as cities or regions have different characteristics in terms of multiple dimensions. This includes socio-demographic information (e.g., age, gender or birthplace), (mobility) infrastructure or environmental data. The aim is to identify, analyse, and visualize relevant data sources for selected test regions.
Keywords:
smart city, urban data, data analytics, data mining, visualization
Thema:
Controlled Image Generation
Art der Arbeit:
Bachelor, Master, Guided Research
Betreuer:
Beschreibung:
With the advent of Generative Adversarial Networks (GAN), image synthesis applications have seen major breakthroughs in recent years. For example, we can generate high-resolution human faces today. However, being able to control the image generation process is important for many practical applications. As such there are multiple ways to gain control over the image generator (e.g., conditioning on labels, layout, text descriptions or learning disentangled representations). Generating images from textual descriptions such as "a red bus on a busy road" offers a flexible and intuitive way for conditional image synthesis, while a layout (bounding boxes / masks + labels) offers fine-grained positional control over the generated objects. There are many possible research directions I can offer within this topic such as experimenting with new architectures, tasks and training regimes, analysing text encoders and embeddings, as well as, investigating new evaluation techniques. Feel free to reach out if the topic sounds interesting or if you have ideas related to this work. We can then brainstorm a specific research question together.
Keywords:
deep learning, gans, text-to-image, layout-to-image, representation learning
Thema:
Exploring Self Attention in Transformers
Art der Arbeit:
Master, Project, Seminar
Betreuer:
Tushar Karayil Ravindran Karayil
Beschreibung:
Transformers have shown a promising new direction in Language generation replacing recurrent neural networks. Different attention mechanisms have been attributed as possible cause for successful performance of these architectures. The goal of this topic would broadly be to develop on and improve the self attention mechanism used in the transformer architectures.
Keywords:
transformers, language generation, nlp
Thema:
Efficient & data type independent CUDA kernels
Art der Arbeit:
Bachelor, Master, Guided Research, HiWi
Betreuer:
Beschreibung:
With typical GPU code all operands must have the same data type. However, the speed of most operations on GPUs is limited not by computation, but by memory bandwidth and latency. In these cases using the smallest possible data types can greatly enhance performance. The standard approach of compiling one version of a kernel for each combination inputs is infeasible though, since the number of combinations, especially for 3 or more inputs, leads to exorbitantly large binaries and very long compilation times. Much more desirable is a flexible kernel that can load inputs of arbitrary data type and converts them to an intermediate representation that can be used to perform the actual computation.
Keywords:
gpu, cuda, efficiency
Thema:
Explainable methods for computer vision and classification
Art der Arbeit:
Bachelor, Master, Project, Guided Research, Seminar
Betreuer:
Beschreibung:
Train or evaluate CNNs under different conditions to induce or measure high-level priors related to explainability.
Keywords:
xai, cv, ml, deep learning, cnn
Thema:
Depth Estimation from Spherical Stereo Images
Art der Arbeit:
Master
Betreuer:
Beschreibung:
3D reconstruction is one of the most exciting but also challenging topics in Computer Vision. It is a fundamental aspect of many innovative technologies. These technologies can be further enhanced by using spherical images, which capture the entire visible scene from a single point in space. This master thesis aims at: 1. investigate the performance of existing depth estimation models for perspective images when applied to spherical images; 2. propose improvements for high resolution spherical images.
Keywords:
dense 3d reconstruction, spherical images, depth estimation, deep learning, cnn
Thema:
Learning Analytics in Education
Art der Arbeit:
Master, Project, HiWi
Betreuer:
Beschreibung:
None
Keywords:
cognitive state, affective state, deep learning, machine learning
Thema:
Natural Language Genration
Art der Arbeit:
Project, Seminar, Master
Betreuer:
Beschreibung:
Enabling deep models to understand visual questions, their rephrasings, or any new task often involves enormous training datasets for better generalization at test times, which evidently is an expensive and time-consuming process. However, by employing existing dataset features, pre-training methods, one can find automatic ways to generate more samples and perform well on multiple tasks. The goal of this research is to investigate such features and devise an approach to building low-resource natural language generation and understanding systems.
Keywords:
natural language processing, natural language generation, visual question answering
Thema:
Self-(re)organizing and especially Forgetful Personal Knowledge Assistants
Art der Arbeit:
Bachelor, Master, Project, Seminar, Guided Research, HiWi
Betreuer:
Beschreibung:
To counter the overwhelming mass of information knowledge workers face in their daily lives, we investigate measures inspired by human memory and cognition, especially forgetting. We are working towards a self-(re)organizing and especially forgetful personal knowledge assistant to support users in their information management and knowledge work activities. This is a "broad" topic comprising several aspects of artificial intelligence. Thus, various theses, projects, etc. with very different foci are possible (please see keywords for details).
Keywords:
context-sensitive assistance and user interfaces, self-(re)organization and digital forgetting, information extraction and retrieval, especially with (near-)real-time constraints, semantic technologies and especially semantic desktop, user activity tracking, data mining, pattern recognition, machine learning
Thema:
Spherical SuperPoint
Art der Arbeit:
Master, HiWi, Project
Betreuer:
Beschreibung:
Interest point (or keypoint) detection is a key problem in Computer Vision. Recent advances in this topic show promising results using CNNs. However, these networks take into account only perspective images and ignore both the potential and growth in popularity of spherical images, which are more suited to a variety of different applications. Motivated by the paper “SuperPoint: Self-Supervised Interest Point Detection and Description”, the focus of the thesis is to extend this approach to spherical images, possibly proposing improvements to the original SuperPoint architecture.
Keywords:
keypoint detection, spherical images, deep learning, cnn
Thema:
Spherical SuperGlue
Art der Arbeit:
Master, Project, HiWi
Betreuer:
Beschreibung:
Robust interest point (or keypoint) matching is essential for Computer Vision applications that involve a moving camera. Recent advances in this topic show promising results using graph neural networks. However, these networks take into account only perspective images (narrow field of view) and ignore both the potential and growth in popularity of spherical images (full 360-degrees field of view), which are more suited to a variety of different applications. Motivated by the paper “SuperGlue: Learning Feature Matching with Graph Neural Networks”, the focus of the thesis is to extend this approach to spherical images, possibly proposing improvements to the original SuperGlue architecture.
Keywords:
keypoint matching, spherical images, deep learning, graph neural networks
Thema:
Anomaly detection in time-series
Art der Arbeit:
Project, Master
Betreuer:
Beschreibung:
Working on deep neural networks for making the time-series anomaly detection process more robust. An important aspect of this process is explainability of the decision taken by a network.
Keywords:
cnn, explainability
Thema:
Bibliographic Reference Detection from Scientific Publications
Art der Arbeit:
Bachelor, Master
Betreuer:
Beschreibung:
Bibliographic reference detection from scientific publication is a challenging task due to diversity in referencing styles and document layout. The purpose of this thesis is to investigate the performance of existing segmentation and object detection models for the task of bibliographic reference detection from scanned publications. Additionally, the extension of existing image-based bibliographic reference detection dataset is also included in the scope of this thesis..
Keywords:
deep learning, cnn, semantic segmentation, object detection, instance segmentation
Thema:
Knowledge-based Deep Learning
Art der Arbeit:
HiWi, Project, Master
Betreuer:
Beschreibung:
None
Keywords:
gnn, graph neural networks, gcnn, graph convolutional neural networks, nlp, natural language processing, word embeddings, graph embeddings
Thema:
Debiasing Datasets for Medical Image Classification
Art der Arbeit:
HiWi, Master, Guided Research
Betreuer:
Beschreibung:
High quality medical image data is rare and acquired in few labs under specific settings. This makes it prone to bias, leading to high classification performance based on the learning of artefacts in the data. Aim of this project is to find and evaluate new ways of debiasing datasets to efface those systematic artefacts while still retaining as much valuable classification-relevant information as possible.
Keywords:
cnn, deep learning, medical image analysis, datasets, preprocessing
Thema:
AI-based Skin Disease Classification from Clinical Images
Art der Arbeit:
Master, Guided Research, HiWi
Betreuer:
Beschreibung:
Naked-eye analysis of skin conditions reveals comparatively less information than other investigation methods. However, modern Deep Learning methods have the ability to utilize provided information more efficiently than humans can. This project aims at training state-of-the-art classifiers for skin disease classification from clinical images and investigating the plausibility of the utilized evidence.
Keywords:
cnn, deep learning, medicine, dermatology
Thema:
Time Series Forecasting Using transformer Networks
Art der Arbeit:
Guided Research, Project
Betreuer:
Beschreibung:
Transformer networks have emerged as competent architecture for modeling sequences. This research will primarily focus on using transformer networks for forecasting time series (multivariate/ univariate) and may also involve fusing knowledge into the machine learning architecture.
Keywords:
time series forecasting, transformer networks
Thema:
Self-supervised Video Object Segmentation
Art der Arbeit:
Master
Betreuer:
Beschreibung:
Exploring the potential advantages of integrating global context for self-supervised video object segmentation.
Keywords:
self-supervision, video object segmentation
Thema:
Attention for video object segmentation
Art der Arbeit:
Master
Betreuer:
Beschreibung:
Exploring RL-based methods for temporal attention in video object segmentation
Keywords:
reinforcement learning (rl), video object segmentation
Thema:
Evaluation of Attribution Methods in the Context of Time-series Analysis using Deep Neural Networks
Art der Arbeit:
Project, Guided Research, Seminar
Betreuer:
Beschreibung:
This work covers the evaluation of different attribution methods used in the context of time-series using deep neural networks. The topic does not only include their applicability and limitations to time-series data. Furthermore, it covers the availability of suitable datasets and comparison metrics to evaluate their performance. As a seminar, this topic covers only the comparison of available methods and datasets. As a project / guided-research, it covers an experiment series to compare the approaches and the collection/creation of suitable datasets.
Keywords:
dnn, cnn, attribution, time-series, explainability
Thema:
Construction & Application of Enterprise Knowledge Graphs in the E-Invoicing Domain
Art der Arbeit:
Bachelor, Master, HiWi, Guided Research Project
Betreuer:
Beschreibung:
In recent years knowledge graphs received a lot of attention as well in industry as in science. Knowledge graphs consist of entities and relationships between them and allow integrating new knowledge arbitrarily. Famous instances in industry are knowledge graphs by Microsoft, Google, Facebook or IBM. But beyond these ones, knowledge graphs are also adopted in more domain specific scenarios such as in e-Procurement, e-Invoicing and purchase-to-pay processes. The objective in theses and projects is to explore particular aspects of constructing and/or applying knowledge graphs in the domain of purchase-to-pay processes and e-Invoicing.
Keywords:
knowledge graphs, linked data, semantic web, knowledge services
Thema:
Complementary Filters in Convolutional Neural Networks
Art der Arbeit:
Bachelor, Master
Betreuer:
Beschreibung:
Convolutional Neural Networks (CNN) made it possible to solve various tasks in the field of machine learning. However, the increasing performance of the current state-of-the-art architectures comes together with the increasing complexity of models. Practioners have noticed that some of the filters in the trained CNNs look complementary to each other. In this project, student has to design and implement the enhanced convolutional layer that defines complementary pairs of filters, and, thus, reduces the amount of trainable parameters.
Keywords:
cnn, deep learning, efficiency
Thema:
Graph Extraction from Printed/Handdrawn Circuit Diagrams/ Piping&Instumentation Diagrams, Generation of Hand-Drawn Diagrams, Interactive Circuit Detection and Simulation
Art der Arbeit:
Bachelor, Master, Project, Seminar, HiWi
Betreuer:
Beschreibung:
I am currently supervising students in multiple topics related to the extraction and simulation of Graph-Based Engineering Drawings from optical sources.
Keywords:
cnn, dynamic system simulation, cv
Thema:
Deep Self-organizing feature maps in time series analysis
Art der Arbeit:
Master, Guided Research
Betreuer:
Beschreibung:
Deep Self-Organising Feature Maps (DSOFM) have shown that they are capable of capturing high-dimensional topologies and are furthermore suitable for image classification on a limited scale. The aim of this project is to apply a stacked version of DSOFM to data from the time analysis domain and to test its suitability for anomaly detection. In addition, advantages and disadvantages, as well as possible improvements, are to be highlighted.
Keywords:
self-organizing feature maps, sofm, unsupervised training, deep learning, time series analysis, python
Thema:
Understanding and enhancing model robustness against adversarial attacks
Art der Arbeit:
Master
Betreuer:
Muhammed Shoaib Ahmed Siddiqui
Beschreibung:
None
Keywords:
deep learning, cnn, robust models, adversarial training, visual recognition
Thema:
Knowledge Graph Construction in Multiple Iterations
Art der Arbeit:
HiWi, Master, Project
Betreuer:
Beschreibung:
Building knowledge graphs from data from scratch is a non-trivial task, especially when we include user feedback in the building process. Because we are most likely not finished after the first iteration, multiple loops have to be performed, to construct the knowledge graph sufficiently. This topic is about a methodology that combines incremental data imports, automated bootstrapping, recurring user feedback loops and knowledge graph updates.
Keywords:
knowledge graph, information extraction, human-in-the-loop, etl, semantic web, rdf
Thema:
Knowledge Graphs für das Immobilienmanagement
Art der Arbeit:
Bachelor, Master
Betreuer:
Beschreibung:
Das Management von Immobilien ist komplex und umfasst verschiedenste Informationsquellen und -objekte zur Durchführung der Prozesse. Ein Corporate Memory kann hier unterstützen in der Analyse und Abbildung des Informationsraums um Wissensdienste zu ermöglichen. Aufgabe ist es, eine Ontologie für das Immobilienmanagement zu entwerfen und beispielhaft ein Szenario zu entwickeln. Für die Materialien und Anwendungspartner sind gute Deutschkenntnisse erforderlich.
Keywords:
knowledge graph, ontologie, corporate memory