Marcel Kollovieh

PhD Student in the Data Analytics and Machine Learning (DAML) group at TUM.

prof_pic.webp

Hi! I am Marcel, a PhD student in informatics at the Technical University of Munich (TUM) under the supervision of Prof. Dr. Stephan Günnemann.

Research: My primary focus is on generative models for non-i.i.d. data, including diffusion models, flow matching, and Bayesian sample inference. I work on time series, graphs, hierarchical structures, molecular generation, and recommender systems. For a comprehensive overview of my background and research projects, please refer to my CV and publications.

Previously: I received my bachelor's and master's degrees in computer science from TUM, and worked as a Student Researcher at Google in Mountain View and as an Applied Scientist Intern at Amazon in Berlin.

Research focus Diffusion models Flow matching Time series Graphs
Google
2026
TUM School of CIT
2023 - Present
Amazon logo
Amazon
2022 - 2023
NUS Logo
NUS School of Computing
2018 - 2019
TUM Department of Informatics
2016 - 2022

News

Our preprint Interpolating Discrete Diffusion Models with Controllable Resampling is now available.
I joined Google as a Student Researcher in Mountain View, working on generative recommendation.
Discrete Bayesian Sample Inference for Graph Generation was accepted at ICLR 2026.
TreeGen, a Bayesian generative model for hierarchies, was accepted at NeurIPS 2025.
Efficient time series processing through token merging was accepted at ICML 2025.
Point set diffusion for unlocking point processes was accepted at ICLR 2025.
At ICLR 2025, we presented flow matching with Gaussian process priors for probabilistic time series forecasting.
Score-based adversarial image generation was published in Transactions on Machine Learning Research.
Expected Probabilistic Hierarchies was accepted at NeurIPS 2024.
Unconditional diffusion models for time series forecasting were accepted at NeurIPS 2023.
In June 2023, I started my PhD in Informatics at TUM’s Data Analytics and Machine Learning (DAML) group under the supervision of Prof. Dr. Stephan Günnemann, working on generative models for non-i.i.d. data.