Session 1: Intro to Inference, ML and Supervised Learning **Location:** MPP, Alps room (A.1.01/03) **Directions:** Main auditorium located on the first floor of the Max Planck Institute for Physics (MPP). After entering the building via its main entrance on the Boltzmannstrasse, please take the stairs to the right of the Heisenberg cafe to get to the first floor. When reaching the first floor, the main auditorium is directly on the right-hand side and facing the Boltzmannstrasse. **Plan:** - Welcome + logistics - Intro AI, ML, and deep learning - Inference, conditional probability, Bayes Theorem, priors - Covariance and Gaussian PDFs - Linear regression and supervised learning

Session 2: Intro to NN and optimizers **Location:** MPP, A.1.01 Alps Sued - Neural networks, MLPs - Optimisation, MSE, convexity - Regularisation - Compression - Neural ODEs

_

Lecture 3: Generative models Generative models - What is a generative model? - Manipulating PDFs for generative models - Simulation-based inference - Normalising flows

Lecture 4: Normalising flows Normalising flows - Change-of-variables models - Analytic flows: Gaussian to Gaussian - Density estimation for Bayesian inference - Neural ODEs recap - Continuous normalising flows

Lecture 5: Continuous-time generative models Continuous-time generative models - Diffusion processes - DDPM - Flow matching - Flow matching issues and cutting-edge methods

_

• Malin Horstmann (TUM)
• Matthias Vigl (TUM)
• Richard Fuchs (ODSL)

Session 6: Automatic Differentiation

• Nicole Hartman (TUM)

**Location:** MPE seminar room https://maps.app.goo.gl/28DAHrHa3AKLWXa66 The magic sauce that makes ML work

Session 7: Classifiers 101

• Nicole Hartman (TUM)

**Location:** MPE seminar room - Perceptron - Multi-layer perceptron / Universal Approximation Theorem - Softmax & multi-class classification - Pytorch in a nutshell

Session 8: Convolutional Neural Networks

• Nicole Hartman (TUM)

**Location:** MPE seminar room

Session 9: Deep Sets

• Nicole Hartman (TUM)

**Location:** MPE seminar room **Intro problem:** [Perceptron-catastrophe notebook][1] **Lecture topics:** - Permutation invariant architectures - Training tricks: regularization [1]: https://gist.github.com/nhartman94/d2c37cb9c28742f4757043180549db92

Session 10: Learning to cluster

• Nicole Hartman (TUM)

**Location:** MPE seminar room Some classical unsupervised learning methods, including: - k-means - PCA

Session 11: Transformers

• Nicole Hartman (TUM)

**Location:** MPE seminar room **Further learning materials:** - [Let's build GPT from scratch][1], Andrea Karpathy - Accompnying [notebook][2] [1]: https://www.youtube.com/watch?v=kCc8FmEb1nY [2]: https://colab.research.google.com/drive/1JMLa53HDuA-i7ZBmqV7ZnA3c_fvtXnx-?usp=sharing#scrollTo=h5hjCcLDr2WC