Machine learning research
Hey, I'm Sascha.
I work on machine learning for tabular data. I'm particularly interested in advancing tabular foundation models and developing methods that people can use and build on.
I became interested in tabular data during my PhD at the University of Mannheim, which I completed in May 2025. Much of my work focused on bringing decision trees and deep learning closer together. After finishing my PhD, I joined TU Clausthal, where my focus shifted towards tabular foundation models, building on that interest in deep learning for tabular data.
I've always wanted to do research with impact. Research that people can use and build on. I enjoy working on ambitious ideas with people who are just as excited about them as I am. With tabular foundation models, there's still so much to figure out, and the chance to help shape what comes next is incredibly exciting.
Selected research & software
Research you can build on.
GRANDE
GRANDE: Gradient-Based Decision Tree Ensembles for Tabular Data
GRANDE brings gradient-based optimization to decision tree ensembles for tabular data. It jointly learns hard, axis-aligned splits, leaf values, and instance-wise leaf weights, combining tree structure with neural optimization.
The PyTorch package provides classification and regression estimators with a scikit-learn-style interface, built-in preprocessing, and CPU/GPU support. Example notebooks cover training and tuning; inspection tools expose individual trees, feature importances, instance-wise estimator weights, and readable rules.
GradTree
GradTree: Learning Axis-Aligned Decision Trees with Gradient Descent
A Python implementation for training hard, axis-aligned decision trees with gradient descent. The package includes preprocessing when categorical feature indices are supplied, a minimal example notebook, and code for reproducing the paper’s experiments.
SYMPOL
Mitigating Information Loss in Tree-Based Reinforcement Learning via Direct Optimization
Code for directly learning interpretable tree policies with PPO. The repository includes a decision-tree actor and neural-network critic, scripts for Gymnasium and vectorized gymnax environments, and experiment commands and hyperparameter configurations used in the paper.
Beyond research
Teaching & mentoring.
Besides research, I really enjoyed giving lectures, supervising students, and co-supervising PhD projects during my time in academia. Something I hadn't expected to enjoy quite that much. The research was fun, but the people made it special.
View my CV ↗Get in touch.
For questions about my research or open-source projects, reach me at sascha.marton@gmail.com.