Pen and paper exercises in machine learning

This is a collection of (mostly) pen-and-paper exercises in machine learning. Each exercise comes with a detailed solution. The following topics are covered:

• linear algebra
• optimisation
• directed graphical models
• undirected graphical models
• expressive power of graphical models
• factor graphs and message passing
• inference for hidden Markov models
• model-based learning (including ICA and unnormalised models)
• sampling and Monte-Carlo integration
• variational inference

A compiled pdf is available on arXiv.

Please use the following reference for citations:

@TechReport{Gutmann2022a,
  author      = {Michael U. Gutmann},
  title       = {Pen and Paper Exercises in Machine Learning},
  institution = {University of Edinburgh},
  year        = {2022},
  arxiv       = {https://arxiv.org/abs/2206.13446},
  url         = {https://github.com/michaelgutmann/ml-pen-and-paper-exercises},
}

The work is licensed under a Creative Commons Attribution 4.0 International License.

Usage

Under linux, you can compile the collection with make. To remove temporary files, use make clean.

By default, the compiled document includes the solutions for the exercises. To compile a document without the solutions, comment \SOLtrue and uncomment \SOLfalse in main.tex.

Contributing

Please use GitHub's issues to report mistakes or typos. I would welcome community contributions. The main idea is to provide exercises together with detailed solutions. Please get in touch to discuss options. My contact information is available here.

Acknowledgements

The tikz settings are based on macros kindly shared by David Barber. The macros were partly used for his book Bayesian Reasoning and Machine Learning. I make use of the ethuebung package developed by Philippe Faist. I hacked the style file to support multiple chapters and inclusion of the exercises in a table of contents. I developed parts of the linear algebra and optimisation exercises for the course Unsupervised Machine Learning at the University of Helsinki and the remaining exercises for the course Probabilistic Modelling and Reasoning at the University of Edinburgh.