Abschlussarbeiten am HCDS

Sie interessieren sich für datengetriebene Forschung, Künstliche Intelligenz oder digitale Technologien – und möchten Ihr Wissen in einem eigenen Projekt vertiefen?
Dann sind Sie bei uns genau richtig!

Am Hub of Computing and Data Science (HCDS) bieten wir Bachelor- und Masterarbeiten mit interdisziplinärem Bezug zu digitaler Transformation, Data Science, Machine Learning, Natural Language Processing, digitalen Methoden in der Forschung und vielem mehr. Unsere Themen stammen aus aktuellen Projekten, sind praxisnah und häufig an reale Anwendungsfälle gekoppelt – oft auch in Kooperation mit Partnern aus Wissenschaft, Verwaltung oder Gesellschaft.

Offene Themen

Benchmarking and optimizing LVLMs to correlate light and electron microscopes

• The thesis should explore potentials to align light and electron microscopy​
• A basic toolchain exists and data is plenty available​
• Collaboration with CSSB​

Kontakt: Dr. Martin Semmann & Xintong​ Wang

Attitude Polarization Detection in Multilingual Text

• Check our Shared task for Attitude Polarization Detection in Multilingual Text: https://polar-semeval.github.io/​
• Participate in the shared task​
• Explore different text analysis approaches and classification models for polarization.​
• Multilingual, cross-lingual approaches​

Kontakt: Dr. Seid Muhie Yimam

NLP for Mental Health – Low-Resource Setup

• Project: INNOVETH: Innovative Health Intervention for Ethiopian Migrant Domestic Workers in the Middle East​
• NLP for mental health​
  • Sentiment analysis​
  • Emotion analysis​
  • Social media text (chat and voice conversations)​
• Doctor-in-the-loop approach​
• Possible travel to Ethiopia or the Middle East

Kontakt: Dr. Seid Muhie Yimam

ASR for Mental Health – Low-Resource Setup

• Project: INNOVETH: Innovative Health Intervention for Ethiopian Migrant Domestic Workers in the Middle East​
• ASR (Whisper, Kaldi,…)​
  • Speech data collection​
  • Finetuning for low-resource languages​
  • App integration​
• ASR for QA in the mental health domain​
• Possible travel to Ethiopia or the Middle East

Kontakt: Dr. Seid Muhie Yimam

Chest X-Ray Analysis for Tuberculosis and Respiratory Disease Screening – Low-Resource Setup

• Project: INNOVETH: Innovative Health Intervention for Ethiopian Migrant Domestic Workers in the Middle East​
• Chest X-Ray Analysis for Tuberculosis ​
  • More than 300k data points available​
  • Fine-tune from existing models​
  • Evaluation (with expert annotation)​
• Explainable AI and Deployment options​
• Possible travel to Ethiopia or the Middle East

Kontakt: Dr. Seid Muhie Yimam

DATS – Test management in a LLM-driven world

• As a growing solution for corpus-oriented research, the requirements for reliability and reproducibility increase. Thus a proper test management concept needs to be developed and implemented.​
• Critical research infrastructures also utilize LLMs; reproducibility and traceability of results is needed​
• Within the thesis a concepts needs to be designed that specifically bridges the gap of traditional software tests and language models​

Kontakt: Robert Geislinger, Fynn Petersen-Frey & Dr. Martin Semmann

DATS – Conceptualizing UIs for Working with Documents and Underlying Embeddings, etc.

• Since the advent of compelling user experiences on our mobile devices, the requirements of users grow.​
• Research software as expert system falls short in this area, especially for modern solutions for multimodal data that rely on embeddings.​
• Goal of the thesis is to assess the state of the art for user experiences and derive a concepts for an enhanced user experience within DATS​

Kontakt: Tim Fischer & Dr. Martin Semmann

OpenWebUI – Making waiting for LLMs more pleasant

• Literature review (ecommerce) on abandonment of requests – Pirate Metrics​
• Review of waiting time mitigation strategies​
• Create prototypes for study: fork OpenWebUI​
• Study with A/B tests on user satisfaction (percieved waiting time etc.)​

Kontakt: Dr. Martin Semmann & Fynn Petersen-Frey

KIDHHS – Relevant topic extraction from German educational plans

• The educational plans (Bildungspläne) are varied in form and writing, so it is hard to quickly understand communalities and differences​
• Derive core concepts and relations of the educational plans​
• Implement a classifier that links social media postings to the concepts of the educational plans​
• Stretch Goal: conceptualize a cross-federal state collection​

Kontakt: Rudy Alexandro Garrido Veliz & Dr. Martin Semmann

KIDHHS – German right-wing ideology classifier​

• Develop a classifier for right wing topics and sentiments within social media​
• Binary classifier, data already at hand​
• Stretch goal: fine-grained classification types of talking points, focus on anti-semitism​​

Kontakt: Rudy Alexandro Garrido Veliz & Dr. Seid Muhie Yimam

Darüber hinaus freuen wir uns über eigene Ideen – Sprechen Sie uns gern an, wenn Sie ein konkretes Thema mitbringen!