Lukas Rauch
Postdoctoral Researcher/Earth Species Project
I am a postdoctoral deep learning researcher currently working at the Earth Species Project (ESP). Previously, I was a PhD student and researcher with the Intelligent Embedded Systems (IES) group at the University of Kassel in Germany.
My work sits at the intersection of deep learning and animal sound. I build audio foundation models that learn from large amounts of unlabeled and labeled recordings, together with the datasets and evaluation infrastructure needed to tell whether they actually work.
- Self-Supervised Learning
- Pretraining audio transformers without labels. Audio SSL for bioacoustics, and probing methods that reveal what the learned representations really encode.
- Bioacoustics
- Foundation models and large-scale benchmarks that make bioacoustics a serious testbed for machine learning, not just an application of it.
- Active Learning
- Deciding what to annotate when expert labeling is the bottleneck, bringing the models to the application in bioacoustics.
My goal is to establish bioacoustics as a frontier domain for deep learning research while translating state-of-the-art methods into practical tools for conservation and environmental monitoring. I am committed to building infrastructure for reproducible science through unified evaluation frameworks and open-source resources.
August 2026 We co-organized Task 4 — Active Learning for Bioacoustics at the BioDCASE 2026 Challenge, built on terrestrial data from BirdSet alongside the marine ATBFL dataset. The task has now concluded and the results are online.
May 2026 I’ve started a new position as a Postdoctoral Researcher at the Earth Species Project (ESP).
April 2026 Our paper BAT: Better Audio Transformer Guided by Convex Gated Probing was accepted at the International Conference on Machine Learning (ICML) 2026 in Seoul, South Korea.
February 2026 Our paper Unmute the Patch Tokens: Rethinking Probing in Multi-Label Audio Classification was accepted at the International Conference on Learning Representations (ICLR) 2026 in Rio de Janeiro, Brazil.
October 2025 We published the pre-print of our work Unmute the Patch Tokens: Rethinking Probing in Multi-Label Audio Classification, currently under review.
August 2025 Our paper Can Masked Autoencoders also Listen to Birds? was accepted to the Transactions on Machine Learning Research (TMLR) 2025.
June 2025 I conducted the second iteration of my full-day workshop Deep Learning with Humans-In-The-Loop: Active Learning for NLP as part of the BERD Academy by BERD@NFDI.
June 2025 We published the pre-print of our work No Free Lunch in Active Learning: LLM Embedding Quality Dictates Query Strategy Success, currently under review.
April 2025 We published the pre-print of our work Can Masked Autoencoders also Listen to Birds?, currently under review.
February 2025 Our paper BirdSet: A Large-Scale Dataset for Audio Classification in Avian Bioacoustics was accepted as a Spotlight Paper at the International Conference on Learning Representations (ICLR) 2025 in Singapore. Our code base is available here.