Latest Research

• Robustness & Adaptation on ImageNet scale: We demonstrated that robustness estimates of ImageNet-scale architectures can be drastically improved by adapting models with simple domain adaptation methods. Batch norm adaptation yields consistent 5-15% points gains across various model architectures. Self-learning can further improve scores, even for models pre-trained on large amounts of data. We obtain scores as low as 22% mCE on ImageNet-C, 17.4% top-1 error on ImageNet-R and 14.8% top-1 error on ImageNet-A. Check out robusta, our PyTorch library for robustness & adaptation on github: https://github.com/bethgelab/robustness.
• Self-supervised Learning: wav2vec and vq-wav2vec demonstrated the effectiveness of contrastive learning for reducing the need for labeled data in speech recognition models, and a lot of progress in contrastive learning has been obtained in other fields of machine learning. We worked on understanding the effectiveness of contrastive pre-training and found that contrastive learning can invert the data-generating process.
• Tools for neuroscience: Approaches for markerless tracking of animals became invaluable for behavioral recording in neuroscience research. We demonstrate improved ood. performance of pre-trained pose estimation networks and discuss principles, pitfalls and perspectives in our recent Neuron primer.

Events

• Jan 2021: I started as a visiting PhD student at Campus Biotech, EPFL, in Geneva.
• Sep 2020: I joined Amazon Web Services in Tübingen on September 1st as a full-time Applied Science intern, advised by Matthias Bethge, Bernhard Schölkopf and Peter Gehler.
• Aug 2020: Our team at KI macht Schule organized a four day AI & ML bootcamp for students; learn more at KI-Camp.de.
• Feb 2020: Together with fellow doctoral students of the Tübingen AI Competence Center, I am organizing a one-day doctoral symposium in February.
• Feb 2020: I joined the Mouse Motor Lab led by Mackenzie Mathis as an ELLIS PhD student. I’ll be working at the Rowland Institute at Harvard in February and March.
• Jan 2020: I co-organized a course on quantum machine learning with Luisa Eck (LMU Munich) and Lucas Stoffl (TU Munich) at the CdE winter school 2020 in Kaub and Oberwesel, Germany.

Meetings, Workshops & Talks

• May 2021: We talked about self-learning for adaptation at the ICLR workshop on weakly supervised learning (contributed talk).
• Dec 2020: We presented our work on batch norm adaptation at NeurIPS 2020.
• Jun 2020: ICML Workshop on Uncertainty & Robustness in Deep Learning: Oral presentation (5% acceptance), of our paper Improving robustness against common corruptions by covariate shift adaptation.
• May 2020: Invited Talk on self-supervised learning and computational creativity at a workshop on “Creativity 2.0” organized by fellows of the German Academic Scholarship Foundation, Tübingen. (postponed due to COVID)
• Apr 2020: ICLR 2020, poster presentation of our paper vq-wav2vec: Self-Supervised Learning of Discrete Speech Representations
• Feb 2020: 1st Tübingen Robust Learning Symposium (Co-Organizer)
• Jan 2020: I gave a talk about “Data-Driven Modelling using Machine Learning: Potentials in Climate Science” in Berlin
• Jan 2020: Co-Organizer, Quantum Machine Learning Course, CdE Winter School
• Sep 2019: We presented wav2vec: Unsupervised Pre-training for Speech Recognition at Interspeech 2019.
• Dec 2019: Workshop Poster Presentations at NeurIPS 2019, Montreal
• Jun 2018: 2nd Smart Start Student Retreat in Freiburg, with contributions from Smart Start Students and our invited speakers Michael Tangermann, Radoslaw Cichy and Stefan Rotter.
• May 2018: Course on Learning in Brains and Machines at the CdE Pfingstakademie 2018 (spring school).
• Dec 2017: Introductory talk on Unsupervised Learning at 15x4 Munich.
• Aug 2017: Co-Organizer for the Biomodels Retreat near Barcelona, Spain.

Software

• Code for our paper Improving robustness against common corruptions by covariate shift adaptation and Adapting ImageNet-scale models to complex distribution shifts with self-learning available as part of our robusta software package: https://github.com/bethgelab/robustness
• (vq-) wav2vec code and models released as part of fairseq
• PyTorch toolbox for Semi-supervised Adaptive Learning Across Domains (salad) for a fair and reproducible evaluation of current domain adaptation algorithms

Press

• wav2vec is presented at Facebook F8 2019. May 1, 2019.
• Facebook details wav2vec, an AI algorithm that uses raw audio to improve speech recognition. Venturebeat, Nov 5, 2019.

I am always looking for motivated students interested in joining me in the Bethge and/or Mathis lab. If you’re interested in working with me on topics around robustness, domain adaptation, reinforcement learning and self-supervised learning at the intersection of neuroscience and machine learning, please contact me at steffen@bethgelab.org.

I am currently working with:

• Shubham Krishna started as a research assistant in the Bethge lab in April 2020. Shubham works on topics around invariant representation learning and is currently doing a Master’s thesis with Bosch AI. [Paper]
• Jin Hwa Lee joined the Bethge & Mathis labs in August 2020. Jin is building self-supervised representation learning algorithms for analyzing neuroscience datasets. [Paper]

In the past, I also worked with:

• Mert Yüksekgönül (now PhD student in Stanford). Mert worked on 3D lifting approaches for analyzing behavioral data.
• Khushdeep Singh (now Research Engineer at INRIA) joined the Bethge lab as a Master thesis student in April 2020. Khushdeep worked on benchmarking and improving the robustness of reinforcement learning algorithms with self-supervised learning. [Paper]
• Xingying Chen was a summer intern in the Bethge & Mathis labs from August - November 2020, modeling adaptation paradigms in neuroscience using reinforcement learning.
• Jan Hansen-Palmus from the Bringmann group at Uni Tübingen was co-advised by Evgenia Rusak and me. He wrote his Bachelor’s thesis on constrained optimization approaches for pseudo-labeling.

A full and up-to-date list is also available on Google Scholar.