# Generative Learning for Molecular Systems

Heidelberg Physics Graduate Days, Winter Semester 2023/2024

October 9–13, 2023

**Times**: 14:00–17:00

**Place**: INF 227 SR 3.403/404

Lecturer: Prof. Dr. Tristan Bereau, Institute for Theoretical Physics, Heidelberg University

## Course description

Generative learning represents one of the most exciting developments in machine learning (ML). Beyond regression or classification, these models produce data points according to an underlying distribution. We will explore recent developments in the context of molecular modeling and statistical mechanics. The course will not assume prior knowledge of ML, and will instead start with basic Bayesian inference. The topical content will follow with an exposure on molecular representations and physical symmetries. Modern ML architectures will be covered, including variational auto-encoders and generative adversarial networks. Finally, we will explore the use of normalizing flows for molecular simulations and free-energy calculations.

## Tentative course outline

- Basics of machine learning
- Molecular representations:
- Physical symmetries in ML for molecules,
- Graph neural networks

- Introduction to generative learning:
- Maximum entropy and restricted Boltzmann machines
- Generative architectures: VAE and GAN

- Boltzmann generators
- Normalizing flows and free-energy calculations

## Main references

The field of generative learning for molecular systems is young. There is therefore no specific text book we will be using. However, the following resources are useful complementary reading:

- Kevin P. Murphy, Probabilistic Machine Learning: An Introduction, MIT Press (2022), https://probml.github.io/pml-book/book1.html
- Kevin P. Murphy, Probabilistic Machine Learning: Advanced Topics, MIT Press (2022), https://probml.github.io/pml-book/book2.html

## Prerequisites

Linear algebra, Probability and statistics, Statistical physics

## Slides

See this page