The Junior Research Group Graph Machine Learning, led by Dr. Martin Ritzert, investigates how to combine structured data such as graphs and hypergraphs with modern machine learning to build models that reason more effectively over complex relational structures. The group focuses on foundational questions: how to design generative models that respect topology and geometry, how to train neural networks for combinatorial optimization, and generally how to push current machine learning (ML) methods on graphs beyond their limitations. Much of the work is empirical, centered on developing graph‑neural architectures that perform reliably while scaling to real‑world datasets.

Projects

A central project of the group is the development of generative models for graphs and tree‑like structures that also produce realistic 3D positions for each node. This work draws its motivation from biological data, particularly 3D neuron scans whose intricate branching remains difficult for generative models to capture. By creating models that can generate realistic neuronal shapes, the group aims to support studies of connectivity patterns, morphological diversity, and structural principles in the brain. These methods also extend to other branching systems, including lung structures and botanical trees.

A second major line of research explores how message-passing machine learning models, the prominent machine learning methodology for large graphs and which can be interpreted as distributed agents, can learn to solve classical combinatorial problems. This includes learning messaging protocols and thus distributed algorithms for tasks such as graph coloring or independent set. It also investigates when and how learning can combine with traditional algorithmic approaches to boost their performance. Additional exploratory work applies graph‑based machine learning to relational databases, where structural information is central but often underutilized by current ML methods.

Team

Lead

The group is led by Dr. Martin Ritzert, who joined ScaDS.AI Dresden/Leipzig as a Junior Group Leader in February 2026. He studied Computer Science at RWTH Aachen University, completing both his bachelor’s and master’s degrees there before pursuing a PhD in theoretical computer science, during which he began working on graph machine learning and published influential work on the expressiveness of graph neural networks. His subsequent postdoctoral research in Aarhus deepened his engagement with machine learning theory, including advances in boosting methods, while his postdoc in Göttingen expanded his focus toward data science, clustering, and graph learning. It was there that he first worked with 3D neuron scan data.
More about his work is available on his personal website: https://scads.ai/about-us/people/martin-ritzert/

Team Members

Currently no other team members.

Publications

By now, no papers by the junior research group exist.
Publications by Dr. Martin Ritzert can be found in his Google Scholar Profile.