PhD
Theoretical Foundations of Artificial Intelligence at TUM
Technical University of Munich
Faculty of Informatics – I7
Boltzmannstr. 3
85748 Garching bei München
Biosketch
Maximilian obtained his Master’s degree in Mathematics at the Technical University of Munich, primarily focusing on statistics and machine learning. During his studies, he gained practical experience through internships in finance and risk management. Maximilian is interested in interpretable machine learning and the statistical foundations of learning. He is pursuing a PhD in the group of Prof. Debarghya Ghoshdastidar at TUM.
relAI Research
Statistical Guarantees for Interpretable Machine Learning
With ML becoming a driving force in many industries, it is crucial that we understand the patterns that complex models learn, and enhance their transparency. I work on two aspects of this:
1. Deriving new algorithms for interpretable machine learning, with provable performance guarantees.
2. Analyzing the representations learned in self-supervised neural networks, the backbone of foundation models.
Let me give an example for each.
1. Clustering is a ubiquitous technique in data science, but rarely interpretable – especially in high-dimensional settings. In my recent ICLR paper (https://arxiv.org/abs/2402.09881), we derive an algorithm that represents nonparametric clusters using a decision tree. Importantly, the method has good theoretical guarantees.
2. Self-supervised learning extracts useful representations from unlabeled data. In our AISTATS paper (https://arxiv.org/abs/2411.11176), we connect SSL to representation learning with kernels. This provides numerous insights.