understanding_ml

Lectures and exercises of my course Understanding Machine Learning at the Karlsruhe Institute of Technology.

There are 9 lectures and 4 exercise sheets.

abstract

Machine Learning is the main driver behind all the rapid progress in the field of Artificial Intelligence over the last decade, prime examples being computer vision and natural language processing. But besides the rather philosophical long-term quest for human-like artificial intelligence, Machine Learning has started to have a much more immediate impact on many aspects of industry, business, and science: as a powerful tool supercharging the scientific method. Progress in learning algorithms and especially increasing computing power and connectivity nowadays allow to make use of the vast amount of data collected by business applications or scientific experiments in order to replace or enhance explicit methods relying on detailed domain knowledge with more generic and automated learning from experience. In this course, we will give an overview of Machine Learning from its foundations to the state-of-the-art methods. For this, we will focus on the explanation of the underlying concepts, like empirical risk minimization or different forms of inductive biases, rather than algorithmic details. We will not only cover the currently dominant methodology of Deep Learning, with its most common techniques like Convolutional Neural Networks or Transformers, but also important „shallow“ algorithms, including Generalized Additive Models or ensemble methods like Random Forests and Gradient Boosting, and show the commonalities between the different algorithmic families. Besides the main focus on supervised learning, we will also touch the most important concepts of Reinforcement Learning and the intersection of Machine Learning and causality.

schedule of lectures

1. introduction
2. statistical learning
3. non-linear models
4. generalization
5. deep learning
6. transformers
7. generative models
8. causality
9. reinforcement learning

schedule of exercises

overarching theme: demand forecasting

1. linear models
2. tree-based methods
3. deep learning
4. quantile predictions and causality