Our teaching connects mathematical foundations, physical reasoning and hands-on computational work. Students should learn not only how to use modern methods, but also how to understand their assumptions, test them experimentally and recognise when they are appropriate for a scientific problem.

We also treat teaching practice itself as something to improve continuously, both pedagogically and in how effectively limited teaching resources are used. Early work included experiments with automated correction and feedback for student submissions, motivated by the need to provide timely and consistent feedback at scale. The aim is to use automation where it improves learning and frees staff time for deeper interaction, while keeping academic judgement and responsibility with instructors.

Current teaching at Cambridge

Sven helped design the MPhil in Data Intensive Science and remains closely involved in shaping the programme as it develops. The aim is to give students a modern preparation for research: rigorous foundations paired with practical experience, increasingly including computational experiments with agentic systems.

M1: Introduction to Machine Learning

This 24-hour major module was designed and is delivered by Sven. It provides a mathematically grounded introduction to machine learning while keeping the connection to scientific applications explicit. The course builds on teaching developed earlier at LMU Munich and has now been delivered twice in Cambridge.

Indicative syllabus

The lecture-by-lecture outline is:

  1. What is machine learning? Course organisation
  2. Fitting a model
  3. The bias–variance trade-off
  4. Optimisers I
  5. Optimisers II and datasets
  6. Probabilistic losses
  7. Linear regression and regularisation
  8. Logistic regression
  9. Multiclass classification
  10. The perceptron
  11. Feedforward neural networks I
  12. Feedforward neural networks II
  13. Regularisation, dropout and hyperparameter optimisation
  14. Support-vector machines I
  15. Support-vector machines II and decision trees
  16. Ensemble methods
  17. Clustering
  18. Gaussian mixture models
  19. Principal component analysis
  20. Autoencoders
  21. t-SNE and UMAP
  22. Energy-based models I: restricted Boltzmann machines and Hopfield networks
  23. Energy-based models II
  24. Evolutionary algorithms and the foundations of reinforcement learning

The precise syllabus may evolve slightly during the term. Assessment is split equally between assessed coursework and a three-hour written examination.

A3: Machine Learning for Particle Physics

This newly designed 16-lecture course is taught jointly with experimental particle physicist Matthew Kenzie, with Sven delivering eight lectures. It explores the unusually close relationship between ideas in particle physics and machine learning, bringing theoretical and experimental perspectives together.

Topics include symmetries as organising principles in particle physics, their incorporation into geometric deep learning, and connections between the probabilistic structures of path integrals, Gaussian processes and deep neural networks. The course treats machine learning and particle physics as mutually informing subjects rather than presenting machine learning only as an off-the-shelf analysis tool.

Student projects

Student projects connect the taught programme to active research. Recent themes range from numerical Calabi–Yau metrics and machine-assisted discovery of dualities to the theoretical foundations of machine learning, including neural tangent kernels and Gaussian processes. Projects combine analytical reasoning with computational experiments and, where appropriate, agentic workflows.

Detailed course materials and current project information are provided to enrolled students through internal University systems.

Previous teaching

At LMU Munich, Sven designed the Machine Learning in Physics specialisation. He created and ran its AI Lab, which brought together researchers from different areas of physics to give students hands-on experience of machine learning in scientific research. He also designed and taught the Master’s course Machine Learning in Fundamental Physics in 2019, 2020 and 2021.

Earlier teaching includes a 24-hour Master’s course on supersymmetry for the MMathPhys at the University of Oxford in 2017, based in part on these lecture notes, as well as student seminars on effective field theory, dualities and string phenomenology.