Statistical Learning (TUM)

Open course materials for MA4802: Statistical Learning at the Technical University of Munich.

Teaching portfolio: Till Richter led the W25/W26 redesign of this MSc module: 11 lecture decks, 12 exercise sheets, and a ~120-page script, from learning theory through diffusion models and flow matching. Fabian Theis (lecturer), Ilia Navosha (co-author). Full credits | Teaching summary

Preview

Lecture 1 (overview)	Lecture 11 (diffusion)	Exercise sheet 1

Download all PDFs from Releases (w25).

People

Role	Person
Lecturer	Fabian Theis
Course lead & primary author (W25/W26 materials)	Till Richter
Co-author	Ilia Navosha
Reviewers	Felix Fischer, Leon Hetzel

Summer 2024 archive: Fabian Theis (lecturer), Till Richter (TA). Details in AUTHORS.md.

Start here

New to the course? Follow the weekly path in Course flow:

1. Open the lecture README for the week (objectives + links).
2. Read outline.md (structure + key bullets) or download the slide PDF.
3. Work through the exercise sheet.md. Solutions are not published here.

No LaTeX installed?

make md          # outline.md (lectures) + sheet.md (exercises)

Browse Markdown in each folder, or use prebuilt PDFs.

Downloads

Source	When to use
GitHub Release w25	Stable semester snapshot (recommended)
GitHub Actions artifact course-pdfs	Latest build from main (90-day retention)
Local make pdf	Contributors editing sources

PDFs are not committed to git. CI builds them on every push to main and attaches them to releases.

Repository layout

style/                 Shared LaTeX theme, macros, and document class
Script/                Course script (companion notes, ~120 pages)
Lecture/
  Images/              All slide figures (shared across lectures)
  Lecture1..11/        Beamer slides, README, outline.md
Exercise/
  Exercise1..12/       Problem sheets (no solutions), README, sheet.md
Archive/               S24 snapshot (Fabian Theis / Till Richter)
scripts/               Markdown export, README generator
.github/workflows/     CI: build all PDFs

Each lecture and exercise folder has a README with objectives, prerequisites, and links.

Course flow

Calendar weeks, lecture folders, and exercise folders do not always share the same number (week 3 still uses Lecture 2). This table is authoritative.

Week	Lecture	Exercise sheet
1	Introduction & learning theory	Bayes classifier
2	kNN, trees, ensembles	kNN, bias-variance
3	Lecture 2 continued	Trees, ensembles, calculus
4	Linear regression	Linear regression
5	Optimization	Optimization
6	Logistic regression	Linear classification
7	Deep neural networks	DNNs
8	Generative vs discriminative	Generative classification
9	Unsupervised I: GMM & EM	EM algorithm
10	Unsupervised II: PCA & SSL	PCA, PPCA
11	Generative I: VAEs	Variational inference
12	Generative II: Diffusion & flow matching	Flow matching

Indexes: lectures | exercises | script

How to use this repo

Goal	What to open
Structure + key points before class	Lecture/LectureN/outline.md
Work problems without LaTeX	Exercise/ExerciseN/sheet.md
Deep reading alongside slides	Script/stat-learning.pdf (Release)
Present or print slides	Lecture PDF from Releases or make lecture-N
Implement XOR / small MLP	Exercise 7 notebook

Weekly workflow: lecture README -> outline or PDF -> script chapter -> exercise sheet.

About outline.md

Lecture outlines are reading guides: sections, slide titles, and main bullet points. Figures and layout stay in the PDF.

Building PDFs locally

Requires latexmk. Package list: tex-packages.txt.

make check-tex
make pdf
make md

See style/README.md for shared LaTeX assets.

Notes

• Solutions and exams are not included in this public repository.
• Archive/S24/ is the Summer 2024 snapshot; active W25/W26 materials are in Lecture/ and Exercise/.
• License: CC BY 4.0. Credits: AUTHORS.md.

Manual build (without Makefile)

Lecture (from Lecture/LectureN/):

TEXINPUTS=../../style//:$TEXINPUTS pdflatex -interaction=nonstopmode "N. Title.tex"

Script (from Script/):

TEXINPUTS=../style//:$TEXINPUTS latexmk -pdf stat-learning.tex