DBIS

The aim of this thesis is to evaluate and extend a developing ontology of explainable data principles, an ongoing work aimed at establishing a structured framework for Data Explainability in AI systems. The current version of this ontology is in its early stages, primarily focused on defining key principles of Data Explainability and exploring their role in enhancing trust and transparency in AI. This thesis will build on this preliminary work by refining the ontology’s structure, proposing new principles or dimensions where needed, and assessing its applicability across various AI domains to make it a robust foundation for responsible AI deployment.

The aim of this thesis is to extend the existing latency analysis of a psychomotor feedback engine within our existing MLOps pipeline [1] [2]. Building upon preliminary latency estimations, this thesis will focus on systematically evaluating each processing step in the pipeline, assessing both theoretical and practical contributions to the overall latency and throughput. By modeling and analyzing latency sources, the goal is to propose and validate optimization strategies that can improve real-time performance for sensor-based AI applications. A particular emphasis will be placed on the throughput of parallel data processing within the infrastructure to ensure timely and efficient feedback delivery.

This master’s thesis builds on the Psychomotor Feedback Engine (PFE) and the IMPECT framework [1], [2], [3], aiming to improve the graphical user interface that teachers use to implement rules and feedback elements in psychomotor learning.

This thesis aims to address those usability issues by redesigning the teacher interface and integrating the IMPECT framework to introduce a more comprehensive feedback mechanism. The IMPECT framework introduces two core concepts: “feedback cards” and exercise stages. Feedback cards provide structured guidance on how exercises should be performed, while exercise stages break down training into a series of sequential tasks. Together, these elements can create a more dynamic and responsive feedback system, offering learners timely, actionable feedback during their training sessions.

This thesis focuses on developing a Transcriptome-Language Model (TLM) to effectively bridge the modality gap between transcriptomic data and natural language text. You will explore advanced models for transcriptomic data representation, and cross-modal learning techniques aligning transcriptomic and textual modalities. This model will be evaluated in tasks such as zero-shot cell property classification and text generation.

This thesis is co-supervised by Sikander Hayat and Rafael Kramann, Department of Medicine II, University Hospital Aachen.

Please send your application to Yongli Mou, M.Sc. (mou@dbis.rwth-aachen.de) and CC. Dr. Sikander Hayat (shayat@ukaachen.de)

This project aims to develop a comprehensive knowledge graph that represents German law documents, including cases and statutes. By creating an ontology tailored to the legal domain and leveraging automated annotation techniques, the project will transform unstructured legal text into structured data that can be queried. This knowledge graph will support legal research, enhance information retrieval, and enable semantic analysis of German legal documents.

This thesis investigates the application of federated learning (FL) to the Personal Health Train (PHT) paradigm, exploring how FL can be better adapted to improve privacy-preserving data analysis in healthcare. The research examines how PHT can facilitate secure, distributed machine learning on sensitive medical data across different institutions, while ensuring data privacy and compliance with regulatory standards.

This research investigates hallucination in vision-language models, focusing on the role of the attention mechanism in contributing to and potentially mitigating hallucinations. The work explores how attention layers influence the integration of visual and textual information and identifies techniques for reducing the generation of inaccurate or irrelevant outputs. A critical research question is understanding how attention mechanisms can be adjusted or improved to decrease hallucination in vision-language models, thus enhancing reliability in applications like image captioning and visual question answering.