Lecture: Deep Learning for Visual Recognition

Course

  • Lecturer(s):
  • Start: 10.10.2019
  • Dates: Thu. 10:15 - 11:45, Meckenheimer Allee 176, Lecture Hall 2
  • Course number: MA-INF 2313
  • Curriculum: Master
  • Effort: 6 CP
  • Exams: 12 February 2020, 12:00-15:00, HS 2 and 17 March 2020, 12:00-15:00 HS 2

Exercises

  • Tutor(s):
  • Start: 24.10.2019
  • Dates: Thu. 16:15-17:45 Seminar Room 3.035b, Endenicher Allee 19A

Description

Neural networks are making a comeback!  Deep learning has taken over the machine learning community by storm, with success both in research and commercially.  Deep learning is applicable over a range of fields such as computer vision, speech recognition, natural language processing, robotics, etc.  This course will introduce the fundamentals of neural networks and then progress to state-of-the-art convolutional and recurrent neural networks as well as their use in applications for visual recognition.  Students will get a chance to learn how to implement and train their own network for visual recognition tasks such as object recognition, image segmentation and caption generation.

No formal pre-requisites.  Students should already be comfortable with concepts in probability theory and optimization and are recommended to have taken at least one course in machine learning or computer vision.  Exercises will be a mix of theory and practical (Python).

News

 

Please inscribe yourself into the mailinglist at: externhttps://lists.iai.uni-bonn.de/mailman/listinfo.cgi/vl-dl
In case you have a problem understanding something, questions related to  exercises/projects, please always feel free to write to the mailing list. This should be a place where you students can talk freely about the lecture, so please do not hesitate to ask and reply!

Exercises

For exercise submissions and course projects, each group is required to have a minimum of 3 students and a maximum of 4. Submission from groups with fewer than 3 students will not be graded.



The exercise sessions will be held on Thursdays as mentioned above (there will be 5 more of them - a total of 6 exercises). Any exception will be notified via the mailing list.

You can submit the theoretical exercises via email to the tutor or to the lecturer during the lecture hours. The solutions will be made available as handouts which you can pick up from Room 3.041. There will be no session explaining the solutions - if you have doubts, please reach out to the tutor. 

For the programming exercise, at least one member from each group is required to come and present the solution. If there is a schedule conflict, please write an email to the tutor for an appointment. There is no need to email the code to the tutor.

 

 

Additional Documents

Slides

Assignment Sheets

Exercise 1: MLBasics
Assignment sheet  (PDF document, 170 KB)
Exercise 2: MLP
Assignment sheet  (PDF document, 125 KB)
Exercise 3: Optimization Regularization
Assignment sheet  (PDF document, 101 KB)
Exercise 4: CNN
Assignment sheet  (PDF document, 117 KB)