DD3364 Elements of Statistical Learning 9.0 credits
Educational levelThird cycle
Academic level (A-D)D
At present this course is not scheduled to be offered.
After successfully taking this course you will
- have a thorough overview and understanding - from derivation to implementation - of many of the established statistical supervised machine learning techniques (see course textbook for an overview),
- be able to apply and adjust a relevant subset of the techniques to your particular research problems,
- be able to describe a learning algorithm in terms of the trade-off it has made with respect to bias and variance,
- be aware of proper training and testing regimes for supervised machine learning problems with limited labelled training data.
Course main content
There will roughly be one lecture per chapter of the course book. However, the following chapters will be omitted Neural Networks, Undirected Graphical Models and Unsupervised Learning as these topics have better coverage in other courses at KTH and in other books such as the Christopher Bishop book Pattern Recognition. Some of the harder and more obscure details within each chapter will also be omitted. Even with these omission the book is still quite long. Therefore, the course will be split into two parts with potentially a break between the scheduling of the two parts. Here is a more detailed description of the content of the two parts of the course.
Students taking the course should definitely be acquainted with the basics of mathematical statistics, linear algebra and have taken some introductory courses in machine learning.
This course will be open to any PhD student at KTH. Potentially advanced and interested master students could also attend, however, they would have to consult the course leader before doing so. It should also be noted that the course is not an introductory one and will cover a lot of material. Therefore it is really recommended that only students with an appropriate background take the course.
The course will use the book Elements of Statistical Learning (second edition) by Trevor Hastie, Robert Tibshirani and Jerome Friedman, 2009. This is available for download on-line, but it is perhaps recommended that students buy it. Students may also find the book Modern Multivariate Statistical Techniques Regression, Classification, and Manifold Learning by Alan Julian Izenman an insightful companion to the main course book for some of the topics covered. It goes into greater depth on some of the issues.
Josephine Sullivan, e-post: firstname.lastname@example.org
Course plan valid from: Spring 12.