Seminar Data-Science I (Methods)

In this seminar, students perform scientific research, either in the form of a literature review or by conducting a small experiment, or a mixture of both, and prepare a written report about the results. Topics of interest focus around a variety of problems and tasks from the fields of Data-Science, Network Science and Text Mining. 

Previous participation in the courses “Network Science” and “Text Analytics” are recommended.

Expertise: Students will acquire a deep understanding of the research topic. They are expected to describe and summarize a topic in detail in their own words, as well as to judge the contribution of the research papers to ongoing research.

Methodological competence: Students will develop methods and skills to find relevant literature for their topic, to write a well-structured scientific paper and to present their results. 

This seminar will be split into four main topic blocks. Every student will be assigned a research paper from only one of these blocks to work on. Yet, it is expected that students also actively participate in the discussion of papers from other topic blocks after they have been presented.

The four topics we are going to discuss in the HWS 2025 are:

1. Fairness in Graph Neural Networks.  Fairness is not only important when considering Large Language Models (LLMs) and textual data. As Graph Neural Networks (GNNs) are increasingly used in real-world applications, including for recommendations and social analysis, ensuring fairness becomes essential. Research has demonstrated that GNNs can be biased by data, vulnerable to fairness attacks, or limited by architectural choices. We will understand how fairness is defined in graph settings and the methods being proposed to make GNNs more equitable and robust.
   • Are Your Models Still Fair? Fairness Attacks on Graph Neural Networks via Node Injections
   • The Devil is in the Data: Learning Fair Graph Neural Networks via Partial Knowledge Distillation
   • Chasing Fairness in Graphs: A GNN Architecture Perspective
   • Fairness-aware Prompt Tuning for Graph Neural Networks
2. Parameter Selection for LLM Inference. This topic explores the impact of inference configurations—such as temperature, top-k, and top-p sampling—on the output of LLMs. As the deployment of LLMs becomes widespread in academia, industry, and society, understanding how decoding choices shape model behavior is critical for ensuring output quality. The papers examine foundational and state-of-the-art research, including sampling algorithms and contrastive decoding methods. It discusses challenges such as hyperparameter sensitivity, task dependency, and the limitations of current evaluation metrics.
   • A Thorough Examination of Decoding Methods in the Era of LLMs
   • DoLa: Decoding by Contrasting Layers Improves Factuality in Large Language Models
   • Turning Up the Heat: Min-p Sampling for Creative and Coherent LLM Outputs
   • Balancing Diversity and Risk in LLM Sampling: How to Select Your Method and Parameter for Open-Ended Text Generation
3. Decision-Making of LLM Agents. Language agents powered by Large Language Models (LLMs) are becoming increasingly important as decision-making tools across a wide range of applications. As their influence grows, so does the need to improve their ability to make accurate, consistent, and context-aware decisions. This project explores various strategies to enhance the decision-making capabilities of LLM agents, focusing on advanced prompting techniques and reinforcement learning approaches. Improving these capabilities is essential to ensure that LLMs can serve as reliable collaborators and support systems in high-stakes and complex decision-making environments.
   • Reflexion: language agents with verbal reinforcement learning
   • Tree of Thoughts: Deliberate Problem Solving with Large Language Models
   • LLMs are Greedy Agents: Effects of RL Fine-tuning on Decision-Making Abilities
   • SELFGOAL: Your Language Agents Already Know How to Achieve High-level Goals
4. Multilinguality. Multilinguality in Large Language Models (LLMs) presents a critical challenge and opportunity as these models increasingly serve users across diverse linguistic backgrounds. While LLMs are often trained on multilingual corpora, their behavior across languages reveals underlying imbalances and biases, often favoring English. We will explore how multilingual language models handle and are influenced by different languages, focusing on cross-lingual data sharing, alignment, and biases.
   • How do languages influence each other? Studying cross-lingual data sharing during LM fine-tuning
   • Do Llamas Work in English? On the Latent Language of Multilingual Transformers
   • Do Large Language Models Have an English Accent? Evaluating and Improving the Naturalness of Multilingual LLMs
   • Steering into New Embedding Spaces: Analyzing Cross-Lingual Alignment Induced by Model Interventions in Multilingual Language Models
      

The schedule below is preliminary, dates are subject to change.

Please register via Portal2.