Hello.A research notebook on Bayesian optimization in high dimensions.
I'm Leonard Papenmeier, a postdoctoral researcher in Machine Learning, currently at the University of Münster, Germany. From September 2020 to June 2025, I obtained my PhD from the University of Lund, Sweden and the Wallenberg AI, Autonomous Systems and Software Program (WASP) under the supervision of Luigi Nardi.
Before that, I completed my Master's degree in Applied Computer Science at the Ruhr-University Bochum, Germany and a Bachelor's degree in Software Engineering from the University of Applied Sciences in Dortmund, Germany.
My work focuses on the optimization of black-box functions with Bayesian optimization, with an emphasis on high-dimensional functions with hundreds of input parameters. I'm interested in exploring the limits of high-dimensional Bayesian optimization and developing scalable, reliable algorithms for a broad range of high-dimensional problems.
Research Output
8 entries · sorted by yearSMOG: Scalable Meta-Learning for Multi-Objective Bayesian Optimization
SMOG is a scalable meta-learning model for multi-objective Bayesian optimization that learns correlations between objectives with a structured joint multi-output Gaussian process, caching meta-task fits and integrating cleanly with standard acquisition functions.
Bencher – Simple and Reproducible Benchmarking for Black-Box Optimization
A modular benchmarking framework that isolates benchmarks in containerized environments and exposes them via a lightweight RPC interface to avoid dependency conflicts.
A Unified Framework for Entropy Search and Expected Improvement in Bayesian Optimization
Shows Expected Improvement as a variational inference view of Max-value Entropy Search and proposes VES-Gamma to blend information-theoretic and EI-style strategies with strong empirical results.
Exploring Exploration in Bayesian Optimization
Introduces observation traveling salesman distance and observation entropy to quantify exploration, revealing how acquisition function behavior links to empirical performance and guiding principled design.
Understanding High-Dimensional Bayesian Optimization
We identify why simple Bayesian optimization methods can work in hundreds of dimensions, showing the role of vanishing gradients from Gaussian process initialization and proposing MSR for state-of-the-art performance.
High-Dimensional Bayesian Optimization with Group Testing
Applies group testing ideas to identify active dimensions before optimizing, improving efficiency on synthetic and real-world tasks while surfacing influential parameters.
Bounce – Reliable High-Dimensional Bayesian Optimization for Combinatorial and Mixed Spaces
Introduces nested embeddings for mixed and combinatorial variables to deliver reliable optimization performance across a wide range of high-dimensional benchmarks.
Increasing the Scope as You Learn – Adaptive Bayesian Optimization in Nested Subspaces
Starts optimization in sparse low-dimensional embeddings and expands them while reusing observations, enabling efficient BO in high-dimensional settings.