PhD Student Bio

I am a PhD candidate in probabilistic machine learning at the University of Tübingen, co-supervised by Robert Bamler and Vincent Fortuin. My research focuses on Bayesian Neural Networks (BNNs), where I develop approximate Bayesian inference methods for uncertainty quantification in deep learning. Specifically, I work on techniques to specify informative function-space priors and scale Bayesian inference to larger datasets and more complex models.

My interests extend to model selection in neural networks and the application of BNNs to decision-making and optimization under uncertainty, including Bayesian optimization, multi-armed bandits, and active learning. I’m also exploring the integration of Bayesian principles into large language models to enhance text generation, calibration, and uncertainty quantification.

I hold a Bachelor’s degree in Communication Systems from EPFL and a Master’s degree in Computer Science from ETH Zürich. During an internship as an Applied Scientist at Amazon in Berlin, I developed Bayesian inference methods for gradient boosting machines, focusing on well-calibrated predictive uncertainty for tabular data with applications to contextual bandit problems.

Through my research, I aim to contribute to the development of robust and reliable deep learning systems, particularly in applications where accurate uncertainty quantification is crucial for decision-making and risk assessment. My goal is to bridge the gap between theoretical advancements in Bayesian deep learning and their practical implementation in real-world scenarios.