Bachelor and Master Theses

We are always looking for motivated students to work on a Bachelor or Master thesis. Below you can find a list of open topics, but please feel free to propose your own topic in any reaearch area of the group.

Open Topics

• Exploring Fuzzy Clustering with Autoencoder for Time Series Data
  This project delves into the utilization of fuzzy clustering alongside autoencoder models for clustering human activity data. By integrating fuzzy clustering with autoencoders, the project aims to improve the accuracy and reliability of classifier predictions by leveraging clustering algorithms to identify and filter confident predictions within the dataset. 

• Evaluating the Impact of Combining Manual Feature Extraction with Raw Data for Training Neural Networks
  This study examines the effectiveness of integrating manual feature extraction techniques with raw data in training neural networks for human activity recognition. By comparing the performance of models trained with and without manual feature extraction, we seek to determine the benefits of incorporating feature engineering into the training process.

• Investigating Methods for Pseudo-Label Selection
  This research focuses on exploring various methods for selecting pseudo-labels in semi-supervised learning scenarios. By evaluating different criteria and strategies for pseudo-label selection, we aim to develop effective techniques for leveraging unlabeled data to improve model performance.

• Comparing Multi-Stage vs. End-to-End Human Activity Recognition
  This project compares the performance of multi-stage and end-to-end approaches for human activity recognition (HAR). By analyzing the strengths and weaknesses of each approach, we aim to identify the most suitable methodology for HAR tasks in different contexts.

• Studying the Effect of Time Inversion in Activity Prediction
  This study explores the feasibility of predicting one activity from another, such as opening and closing doors, by reversing time sequences. By analyzing the temporal relationships between activities, we aim to determine the potential for generating activities from their opposites.

Theses in Progress

• Deep Learning-basierte Analyse von Hämatomen in Hyperspektralbildern

• Process Mining for Human Activity Sequences in Warehouses
  This project focuses on applying process mining techniques to analyze human activity sequences in warehouse environments. By leveraging process mining algorithms and data collected from sensors, the project aims to uncover processes performed in warehouses. The insights gained from this analysis can inform process optimization strategies, improve workflow efficiency, and enhance overall warehouse operations.

• Developing a Semi-Supervised Transformer Model for Activity Recognition
  This project focuses on designing and implementing a semi-supervised transformer model for activity recognition tasks. By leveraging both labeled and unlabeled data, the model aims to achieve improved performance and generalization capabilities compared to traditional supervised approaches.

• Analyse von Schweinswal-Lauten mit Machine Learning-Methoden

• Erkennen und Lokalisieren von Anomalien in Wärmebildern von Industriebauteilen mithilfe unüberwachter Lernmethoden