news

Sep 01, 2026 I have submitted DOODAT (Dynamic Optimisation of Discount and Trace Decay for Value-Based Reinforcement Learning) to ICLR. This work grew out of my research visit at Mila. Fingers crossed!
Aug 01, 2026 I organised the AutoRL workshop at the Reinforcement Learning Conference (RLC) and presented ARLBench at RLC this August.
Jun 01, 2026 Michael Ebbers, whose bachelor’s thesis I supervised, has won the Schöneborn Prize at RWTH Aachen University for his outstanding thesis, “On the Importance of Noise Sources for the Variability of Hyperparameter Performance in Reinforcement Learning”. Congratulations, Michael!
Sep 01, 2025 Our paper, Performance Prediction In Reinforcement Learning: The Bad And The Ugly, was published at the 18th European Workshop on Reinforcement Learning (EWRL 2025) and is available on OpenReview. We investigate the challenges of predicting RL performance from hyperparameter configurations and highlight the limitations of surrogate models on rugged, noisy performance landscapes.
Sep 01, 2025 I have started my ten-month research visit at Mila in Montreal, working with Professor Pablo Samuel Castro on Automated Reinforcement Learning. The visit is fully funded by a DAAD fellowship.
Jan 01, 2025 I am now part of the AI Grid initiative, which aims to strengthen the exchange and synergies between AI Doctorates in Germany.
Oct 01, 2024 Our joint work about the Automated RL Benchmark was accepted at the European Workshop of Reinforcement Learning this year and is now available at Arxiv. The benchmark contains lightning fast implementations of three common RL algorithms in JAX, build specifically to benchmark AutoRL methods in a consistent setting.
Jun 22, 2024 I am co-organising the AutoRL workshop at ICML this year with an amazing team. Very much looking forward to the workshop and to meet all the people excited about AutoRL.
Mar 30, 2024 My first AutoRL paper as a PhD student was accepted at this year’s Reinforcement Learning Conference, where we explored the potential of jointly optimising hyperparameters and reward functions. I am very excited to present our work this August in Amherst!
Apr 16, 2023 Our paper based on the results of my master’s thesis was accepted at this year’s ICCASP Workshop Self-supervision in Audio, Speech and Beyond. We examined different methods in pre-training and fine-tuning to adjust large transformer models for low ressource Automatic Speech Recognition.