Michał Wiliński

First-year PhD student, Robotics Institute, Carnegie Mellon University

About

I am a first-year PhD student at the Robotics Institute at Carnegie Mellon University. I am broadly interested in the problem of learning in modern generative models, especially language models. My focus is on the construction and specification of learning problems (environment, reward, objective, optimization).

In 2024, I was a Robotics Institute Summer Scholar (RISS) at CMU. Since then, I collaborated with the Auton Lab on pretrained time series models and interpretability.

I completed my B.Sc. in Artificial Intelligence with honors at Poznan University of Technology. My undergraduate thesis explored representation-level interventions for mitigating harmful learned biases in vision models. Earlier projects with PUTvision Laboratory covered space SLAM for the LeopardISS mission. I also collaborated with Machine Learning Laboratory at PUT.

Research Interests

I'm interested in how generative models learn from interaction and expert demonstrations. I draw on ideas from imitation learning, reward learning, reinforcement learning, and related areas.

Most recently, I have been working with Liu Leqi and Chirag Nagpal on how inverse reinforcement learning can infer human-aligned reward functions from demonstrations. Restricting the reward to combinations of known evaluators turns the classical reward-identifiability problem from unconstrained function recovery into a lower-dimensional, interpretable problem of estimating how humans trade off among objectives.

Selected Publications

For a complete list, see my Google Scholar profile.

* Equal contribution.

Education

  • Carnegie Mellon University — PhD, Robotics Institute, School of Computer Science, 2026–present.
  • Poznan University of Technology — B.Sc. in Artificial Intelligence, 2021–2025.