The lab works on robust, identifiable and interpretable machine learning tools for scientific inference. We work on both core machine learning topics and the application of machine learning algorithms for scientific inference in the life sciences. Our research focus is on:

  • Representation learning, especially using contrastive learning and generative pre-training
  • Hierarchical representation learning
  • Identifiability and robustness of machine learning algorithms
  • (Nonlinear) dynamical system modeling

The following is a list of all pre-prints and publications published by lab members (including work done at previous institutions).

Learnable Latent Embeddings for Joint Behavioral and Neural Analysis

CEBRA is a contrastive learning algorithm for relating behavior and neural activity.

Contrastive Learning Inverts the Data Generating Process

Contrastive learning with the InfoNCE objective can recover the ground truth latent factors underlying the data.

A Primer on Motion Capture with Deep Learning: Principles, Pitfalls and Perspectives

We discuss principles of pose estimation algorithms and highlight their potential and pitfalls for experimentalists.

Improving Robustness against Common Corruptions by Covariate Shift Adaptation

Adapting batch norm statistics of trained computer vision models considerably increases robustness at test-time.

vq-wav2vec: Self-Supervised Learning of Discrete Speech Representations

Learning discrete representations of speech yields state-of-the-art recognition performance on TIMIT and WSJ.

Iron-Sequestering Nanocompartments as Multiplexed Electron Microscopy Gene Reporters

Multicolored gene reporters for light microscopy are indispensable for biomedical research, but equivalent genetic tools for electron …

wav2vec: Unsupervised Pre-training for Speech Recognition

Contrastive pre-training on speech at scale reduces the need for labeled data.

Multi-Task Generalization and Adaptation between Noisy Digit Datasets: An Empirical Study

Transfer learning for adaptation to new tasks is usually performed by either fine-tuning all model parameters or parameters in the …

Salad: A Toolbox for Semi-supervised Adaptive Learning Across Domains

We introduce salad, an open source toolbox that provides a unified implementation of state-of-the-art methods for transfer learning, …

Towards Desynchronization Detection in Biosignals

This study presents a novel data-driven approach to detect desynchronization among biosignals from two modalities. We propose to train …

Context-based Normalization of Histological Stains using Deep Convolutional Features

While human observers are able to cope with variations in color and appearance of histological stains, digital pathology algorithms …