Selected Problems in Machine Learning
- Teacher: David Mareček
- Time and location: Wednesday 9:00–10:30, S7
The course has been designed for students with a deep interest in Machine Learning. The course is a flexible combination of lectures, discussions, exercises and literature reading, aimed at the following topics:
- Bayesian inference
- practising unsupervised ML (especially methods based on sampling)
Students are expected to be familiar with basic probabilistic and ML concepts, roughly in the extent of NPFL067/068 - Statistical methods in NLP I/II, and NPFL054 - Introduction to Machine Learning (in NLP).
Course passing requirements
- All students are required to actively participate in the classes.
- ~2 homeworks (programming assignments)
- October 4th: Introduction and Warm-up test - let me know what you already know
- October 11th: Answering questions form the Warm-up test, introduction into probability, information, and learning theories.
- October 18th: Beta-Bernouli and Dirichlet-Categorial models
- slides for Beta-Bernouli and Dirichlet-Categorial models by Carl Edward Rasmussen from University of Cambridge
- how to compute expected value of the Beta distribution can be found on YouTube
- web application showing the Beta-Bernouli distribution and many others can be found at RandomServices.com
- October 25th: Modeling document collections, Categorical Mixture models, Expectation-Maximization
- November 1st: Gibbs Sampling, Latent Dirichlet allocation
- November 15th: Assignment - Latent Dirichlet Allocation:
evaluation, document perplexity
- November 22th: Chinese segmentation, Chinese restaurant process
- November 29th:
- December 6th:K-means vs. Gaussian mixture models
- December 13th: Sampling Methods: Rejection Sampling, Importance Sampling, Metropolis-Hastings Sampling
- see Chapter 11 in the book Christopher Bishop: Pattern Recpgnition and Machine Learning
- December 20th: We will only discuss problems with your homework, you do not need to come.
- January 10th: Final test, homework evaluation, tree structures sampling.