Fynn Gerding

Fynn Gerding

PhD researcher studying information, structure, and efficiency in neural networks.

Research

You can also find my articles on my Google Scholar profile.

Theses

Reinforcement Learning on Aperiodic Tilings

M.Sc. thesis, Radboud University, 2026

My M.Sc. thesis extends my B.Sc. work from the Penrose P3 tiling to Penrose P2 and the recently discovered aperiodic monotile, packaged as reinforcement-learning environments in the gymnasium framework (Penrose-P3-v0, Penrose-P2-v0, Einstein-v0). It also develops an algorithm that exploits novelty in Markov chains to improve exploration and navigation. Whereas periodic environments repeat the same structure everywhere, aperiodic tilings have regularities that depend on scale: increasing the agent’s radius of exploration demands more structural information and makes convergence explicitly scale-dependent. The thesis is not publicly available.

Navigating Aperiodicity: Challenges of Reinforcement Learning

B.Sc. thesis, Radboud University, 2024

Reinforcement learning in the Penrose P3 tiling, where the action space varies from tile to tile and the Markov property no longer holds. Using PPO with recurrent and Transformer policies, I study how giving the agent its past trajectory as context, and adding positional embeddings, affects convergence and attention. Advised by Tal Kachman and Serge Thill. Available in the Radboud thesis repository.