About me
I am a final-year Ph.D. student in the Machine Learning group at the University of Kaiserslautern-Landau, supervised by Professor Sophie Fellenz. My research primarily focuses on reinforcement learning and value-aligned decision-making.
I am a big coffee and kombucha enthusiast, and I love discovering new cafés, flavors, and varieties whenever I get the chance.
🐣 Research Interests
- Inference-time value alignment
- Multi-objective decision-making
🐥 Publications
- W.Li, et al. (2026) Pluralistic AI Alignment Requires Inference-Time Multi-Objective Control. Advances in Neural Information Processing Systems (NeurIPS), 2026.
- W.Li, et al. (2026) TORA: Train Once, Realign Anytime for Offline Multi-Objective Reinforcement Learning. Proceedings of the AAAI Conference on Artificial Intelligence.
- W. Li, et al. (2025). Inference-Time Value Alignment in Offline Reinforcement Learning: Leveraging LLMs for Reward and Ethical Guidance. Workshop on WORDPLAY: WHEN LANGUAGE MEETS GAMES at EMNLP.
- W. Li, W. Mustafa, et al. (2025). Inference-Time Preference-Aligned Diffusion Planning for Safe Offline Reinforcement Learning. Proceedings of the Third Workshop on Hybrid Human-Machine Learning and Decision Making (HHMLDM) at ECML-PKDD (Oral Presentation).
- W. Li, R. Devidze, W. Mustafa, and S. Fellenz.(2024) Ethics in Action: Training Reinforcement Learning Agent for Moral Decision-making In Text-based Adventure Games. Proceedings of International Conference on Artificial Intelligence and Statistics (AISTATS)
- W. Li, R. Devidze, S. Fellenz.(2023) Learning to play text-based adventure games with maximum entropy reinforcement learning. European Conference on Machine Learning and Data Mining (ECML-PKDD) Abstract Poster code
- W. Li.et.al.(2022),Topic-Guided Knowledge Graph Construction for Argument Mining. IEEE International Conference on Big Knowledge(ICBK)
📝 Teaching
I supervise Master projects and Thesis. Please contact me if you are interested in the following topics:
- Deep Reinforcement learning with application to chemical data.
- Pre-training LLMs for molecular generation