Michael Petrowski

Hi, I’m Michael. Currently, I am a Visiting Graduate Student at the Computational and Biological Learning (CBL) Lab in the Department of Engineering at the University of Cambridge, supervised by Dr Flavia Mancini. Alongside this research visit, I am pursuing an MSc in Artificial Intelligence and Data Science at Heinrich Heine University Düsseldorf and working as a Student Researcher at the Dialogue Systems and Machine Learning Lab under Prof. Milica Gašić.

My bachelor thesis investigated whether equipping artificial agents with introspection could enhance adaptive learning. I developed a reinforcement learning framework where agents infer internal belief states to dynamically guide exploration. This approach demonstrated that subjective reward shaping significantly improves performance while replicating complex, human-like behavioral dynamics. The resulting paper, Exploration Through Introspection, was presented at the AAAI-26 Workshop on Theory of Mind. At Cambridge, I am expanding this research into LLM post-training by investigating how belief-dependent subjective reward functions can guide language-model generation.

Research interests: deep reinforcement learning, theory of mind, sample-efficient learning, reward design