README.md

Course on Automated Machine Learning (AutoML)

Introduction

Based on machine learning (ML), AI achieved major breakthroughs in the last years. However, applying machine learning in practice is a challenging task and requires a lot of expertise. Among other things, the success of ML applications depends on many design decisions, including an appropriate preprocessing of the data, choosing a well-performing machine learning algorithm and tuning its hyperparameters, giving rise to a ML pipeline. Unfortunately, even experienced ML experts need days, weeks or even months to find well-performing ML pipelines. To support ML users, developers and researchers, automated machine learning (AutoML) was proposed. AutoML tools propose well-performing ML pipelines for a dataset at hand such that the development time of new ML applications will be reduced, the efficiency of the users increases and the required expertise in ML will be minimized. Therefore, AutoML helps to apply ML to new applications and fosters the democratization of machine learning.

Content

• Algorithm Selection
• Hyperparameter Optimization
• Neural Architecture Search
• Learning2Learn
• Dynamic Configuration
• Interpretability of AutoML
• Algorithm Configuration

Prerequisites

The course is targeted at students and practitioners with solid background in machine learning. You should the concepts of

• Hands-on experience in applying ML and DL (the most important prerequisite)
• Basic supervised models, e.g., linear models, SVMs and tree-based models
• Basics in deep learning, e.g., feed forward network, CNN, RNN
• Evaluation of supervised ML, e.g., different metrics, cross validation
• Basics in maths, e.g., linear algebra, calculus, derivatives
• Basics in optimization, e.g., stochastic gradient descent
• Hands-on experience with Python or R

Organizers and Contributors

• University of Hannover, Germany:
  • Marius Lindauer
  • Difan Deng
  • Carolin Benjamins
  • Daniel Ritter
• Ludwig Maximilian University of Munich, Germany:
  • Bernd Bischl
  • Janek Thomas (Fraunhofer)
  • Jakob Richter
  • Julia Moosbauer
  • Omid Charrakh
• University of Freiburg, Germany:
  • Frank Hutter
  • André Biedenkapp
  • Arber Zela
  • Matthias Feurer
  • Guri Zabergja
  • Maciej Janowski
  • Archit Bansal
• University of Wyoming, USA:
  • Lars Kotthoff
• University of Eindhoven, Netherlands:
  • Joaquin Vanschoren

Software Dependencies

slides.py requires pdftk (e.g., run sudo apt install pdftk on a Ubuntu). In addition, it requires Python 3.9 (or later).

License

All content created by us is licensed under a Creative Commons Attribution-ShareAlike 4.0 International License.