Yannick Metz
I am a postdoctoral researcher working on Human-AI Communication and Reinforcement Learning from Human Feedback at ETH Zürich.
Research Interest: Reinforcement Learning from Human Feedback, Human-AI Interaction, Explainable AI, Visual Analytics
I am a research assistant at the ETH IVIA Lab under Prof. Mennatallah El-Assady. Until recently, I was a PhD student at the University of Konstanz, advised by Prof. Daniel Keim. I was also a visiting researcher at Worcester Polytechnic Institute (WPI) in the US in 2024. I work on Human-AI Communication, combining reinforcement learning from human feedback, interactive visualizations for explainability, and uncertainty in RL.
News
- July '26: We will present our work MAVRL at ICML 2026 in Seoul. [ArXiv] [Website]
- July '26: Our work on emotion vectors in open-source LLMs was accepted to the Mechanistic Interpretability Workshop at ICML 2026. [ArXiv]
- June '26: I started as an active contributor to the evals and benchmarking team of Apertus, the open-source LLM from Swiss AI, and Apertus Claritas, a portal for interpretability research outreach.
- Dec '25: Timo Kaufmann presented our work ResponseRank at NeurIPS 2025. [ArXiv] [Code]
My Research
My main research aim is the improvement of bidirectional communication between humans and AI agents for effective learning and alignment. In particular, I focus on the following topics:
- Widening the space of human feedback for RL: Going beyond preference learning, and towards diverse, expressive, contextual feedback.
- Reinforcement Learning from Human Feedback: Integrating different feedback types into the reward learning pipeline.
- Interactive Visualizations for Explainability: Understanding RL agent behavior allows for better feedback and more trustworthy interactions.
- Uncertainty in Reinforcement Learning: Measuring an agent's uncertainty is crucial for effective human-AI communication.
Publications
[1] Raphaël Baur, Yannick Metz, Maria Gkoulta, Mennatallah El-Assady, Giorgia Ramponi, Thomas Kleine Buening MAVRL: Learning Reward Functions from Multiple Feedback Types with Amortized Variational Inference
International Conference on Machine Learning (ICML), Conference Paper, 2026
[2] Sinie van der Ben, Raphaël Baur, Yannick Metz, Mennatallah El-Assady Where Do Models Find Happiness? Emotion Vectors in Open-Source LLMs
Mechanistic Interpretability Workshop at ICML, Workshop Paper, 2026
[3] Timo Kaufmann, Yannick Metz, Daniel Keim, Eyke Hüllermeier ResponseRank: Data-Efficient Reward Modeling through Preference Strength Learning
The Annual Conference on Neural Information Processing Systems (NeurIPS), Conference Paper, 2025
[4] Yannick Metz, András Geiszl, Raphaël Baur, Mennatallah El-Assady Reward Learning from Multiple Feedback Types
International Conference on Learning Representations (ICLR), Conference Paper, 2025
[5] Yannick Metz, David Lindner, Raphaël Baur, Mennatallah El-Assady Mapping out the Space of Human Feedback for Reinforcement Learning: A Conceptual Framework
ArXiv Preprint, Survey Paper, 2024
[6] Yannick Metz, David Lindner, Raphaël Baur, Daniel A. Keim, Mennatallah El-Assady RLHF-Blender: A Configurable Interactive Interface for Learning from Diverse Human Feedback
ICML2023 Interactive Learning from Implicit Human Feedback Workshop, Full Paper, 2023
[7] Yannick Metz, Eugene Bykovets, Lucas Joos, Daniel A. Keim, Mennatallah El-Assady Visitor: Visual interactive state sequence exploration for reinforcement learning
Full Paper (Eurovis), Computer Graphics Forum, 2023
[8] Yannick Metz, Udo Schlegel, Daniel Seebacher, Mennatallah El-Assady, Daniel A. Keim A Comprehensive Workflow for Effective Imitation and Reinforcement Learning with Visual Analytics
Short Paper, EuroVA, 2022
[9] Eugene Bykovets, Yannick Metz, Mennatallah El-Assady, Daniel A. Keim, Joachim M Buhmann How to Enable Uncertainty Estimation in Proximal Policy Optimization
ArXiv Preprint, 2022
[10] Dirk Streeb, Yannick Metz, Udo Schlegel, Bruno Schneider, Mennatallah El-Assady, Hansj{"o}rg Neth, Min Chen, Daniel A. Keim Task-based visual interactive modeling: Decision trees and rule-based classifiers
IEEE Transactions on Visualization and Computer Graphics, 2021
[11] Yannick Metz, Dennis Ackermann, Daniel A. Keim, Maximilian T Fischer Interactive Public Transport Infrastructure Analysis through Mobility Profiles: Making the Mobility Transition Transparent
Visualization in Data Science (VDS at IEEE VIS), 2024
[12] Maximilian C Hartmann, Moritz Schott, Alishiba Dsouza, Yannick Metz, Michele Volpi, Ross S Purves A text and image analysis workflow using citizen science data to extract relevant social media records: Combining red kite observations from Flickr, eBird and iNaturalist
Ecological Informatics
[13] Mennatallah El-Assady, Rebecca Kehlbeck, Yannick Metz, Udo Schlegel, Rita Sevastjanova, Fabian Sperrle, Thilo Spinner Semantic color mapping: A pipeline for assigning meaningful colors to text
2022 IEEE 4th Workshop on Visualization Guidelines in Research, Design, and Education (VisGuides)
[14] Yannick Metz Effective Imitation and Reinforcement Learning Through Visual Analytics
Masters Thesis
[15] Yannick Metz Evaluating Concept Attributions for the Interpretability of Deep Neural Networks
Bachelor Thesis
Code
Top 5 GitHub repositories by stars.
- rlhfblender (Python | 14 stars)
RLHF-Blender: A Configurable Interactive Interface for Learning from Diverse Human Feedback - smartgpt_ui (TypeScript | 9 stars)
UI Implementation for SmartGPT - multi-type-feedback (Jupyter Notebook | 7 stars)
Implementation for the paper: Reward Learning from Multiple Feedback Types (ICLR2025) - rlhfblender-ui (TypeScript | 3 stars)
User Interface Implementation for: RLHF-Blender: A Configurable Interactive Interface for Learning from Diverse Human Feedback - kaggle_arc (Python | 1 stars)
Repostitory containing WIP for Kaggle Abstraction&Reasoning Challenge
Teaching
- TA & Lecturer: XAIML - Interactive Machine Learning: Visualization & Explainability (ETH Zurich) - FS 2026 (Frühlingsemester 2026)
- TA: Data Mining: Basic Concepts (UKON) - WS 2022/23, WS 2023/24, and WS 2025/26
- TA: Seminar: Data Visualization and Analysis (UKON) - WS 2024/25
- TA: Data Visualization (UKON) - SS 2023
- TA: Natural Language Processing/Document Analysis (UKON) - SS 2022
- TA: Programming Course: Object-Oriented Programming (UKON) - WS 2021/22 and SS 2024
Education & Past Work
I have a M.Sc. and B.Sc. in Computer Science from the University of Konstanz, Germany. I have also studied abroad in Uppsala, Sweden for a year. During my studies, I focused on machine learning, particularly explainable AI and reinforcement learning. During my studies, I worked for 2 years as a working student at Airbus Defence and Space in Immenstaad, Germany. I have also worked as a student research assistant at the University of Konstanz. I also worked as a freelance machine learning engineer with Körber Supply Chain Logistics on predictive maintenance, computer vision, and robotics.