Much of our research applies AI and Machine Learning (see further below), but we are also interested in developing, improving or understanding Deep Learning or Statistical Learning/Modeling in general: How does it work? When does it fail? What can we do when data is scarce? Projects may involve re-implementing methods from the literature, systematic experiments with designed data sets, attempts at qualitative or performance improvements, etc. These projects generally require a background in Machine Learning.

Contact (if not stated otherwise): Jan Schlüter

These student projects can be started any time (including holidays) and can span over semester boundaries.

Remark: We are open for new proposals - if you are interested in Machine Learning / Artificial Intelligence, feel free to contact us!

Topics:

Interpretable Machine Learning / Explainable Artificial Intelligence (Contact: Katharina Hoedt):

• Concept-based Explanation Methods (e.g. TCAV, Non-negative Concept Activation Vectors, Concept Bottleneck Models, ...)
• Evaluating Explanations
• Mechanistic Interpretability
• Counterfactual Explanations

Adversarial Examples (Contact: Katharina Hoedt):

• Adversarial Robustness
• Reasons for Adversarial Vulnerability
• Linking Adversarial Vulnerability and Explainability

Representation Learning:

• Deep InfoMax, opens an external URL in a new window
• Invariant information clustering, opens an external URL in a new window
• Disentangled representation learning, e.g..:
  • Challenging Common Assumptions in the Unsupervised Learning of Disentangled Representations, opens an external URL in a new window

Generative Models:

• Generating raw audio GANs / WaveNet
  • Recommended Reading: Generative Models | blog post at openai.com, opens an external URL in a new window
  • WaveNet: A Generative Model for Raw Audio, opens in new window
• DDSP: Differentiable Digital Signal Processing, opens an external URL in a new window
• Flow-based models, opens an external URL in a new window

Machine Learning Theory:

• Understanding deep learning requires rethinking generalization, opens an external URL in a new window