Bartlomiej Pogodzinski

I am a PhD student in the Geometric Representation Learning group at MPI-INF, supervised by Jan Eric Lenssen. Before starting my PhD, I was part of the group as a HiWi and later as a Master's student.

My research focuses on improving generative models — with a focus on fundamentals of flow and diffusion models. During my Master's, I co-authored work on Spatial Reasoning Models (ICML), studying how decision order affects the difficulty of modelling continuous domains, and extended this in my thesis to non-image domains, leading to the workshop paper Spatial Reasoners for Continuous Variables in Any Domain (ICML CODEML). My current PhD work investigates jointly generating images with auxiliary representations learned specifically to facilitate denoising.  

Note to Students Interested in Master's Theses or HiWi Positions

We receive many inquiries each week, and unfortunately, most are overly generic and show little awareness of our group's research. If you're genuinely interested in joining us, please take the time to explore our current work. In your email, highlight how your background aligns with our research interests. Ideally, propose a concrete project idea or a topic you're enthusiastic about. Doing this significantly increases your chances of a positive response.