Background book: http://www.deeplearningbook.org/
Lectures twice a week during 4 weeks and exercises once a week during 4 weeks. The exercises include programming work.
In addition, there is a competition programming project, which you will present during the last week of the course in form of a short paper and presentation about what you have been doing. The competition project is conducted as group work (2-3 persons per group). The groups write code and a short documentation. The competition will have two datasets, of which you choose one to solve with your group. Your group must write all of the code themselves (the exercises will prime you for this). The supervisors will hold a testset, against which your results are compared to, once in the middle of the competition as feedback for you, and once when the competition ends. Leaderboards will be visible for everyone on the course. Your position on the leaderboards will not affect your grading, as long as the results are reasonable (separate benchmarks will be set for both datasets). Grading is based on the final code itself, as well as the documentation, which should reflect the parametrizations you have tried through the competition. We hope to see varied results for the sake of learning experience, so don't worry about getting bad results: everyone learns more from the diversity.
The datasets will be published later.
Evaluation: 20 points for the exercises, 20 points from exams and 20 points from competition.
The course cannot be passed with an exam only.
Please note that the course registration will end 25 October.
In case of overbooking of the course, 34 students will be selected.