Deep learning

If you want to participate in this course, please register via C@mpus.


You can find Deep Learning time schedule for SS19  here.

New 6ECTS Deep learning modul (75960) in SS2019 replaces the old 3ECTS Deep learning modul (77920). But ISS still offers the exam for the old 3ECTS module. This means, in SS2019 there will be two different exams, one for the old 3ECTS and one for the new 6ECTS modul.
For mor information, see here.

Recommended prerequisites

Solid knowledge about matrix computation, probability theory as
well as basic knowledge about optimization as from the course
"Advanced mathematics for signal and information processing" are highly
Knowledge about general methods for pattern recognition as from the
course “Detection and pattern recognition” is recommended.

learning goals

• Learn the basic tasks and concepts of machine learning (density
  estimation, regression, classification, model, representation, …)
• Learn the differences between conventional (shallow) concepts of
  machine learning and deep learning
• Learn the most basic deep architectures (DNN, auto-encoder, CNN,
  RBM, RNN) and issues of training (how to parametrize, initialize and
• Learn to understand and reduce a trained DNN (visualization, model
• Learn how to use Theano, a widespread Python Toolbox for deep
  learning, to implement the introduced deep architectures in Python


• Important basics from statistics (Entropy, cross-entropy, KL-divergence,
  important inequalities)
• Tasks and concepts from machine learning (density estimation,
  regression, classification)
• The most basic deep architectures (DNN, auto-encoder, CNN, RBM,
• How to train a network and to perform inference
• Concepts for visualization and reduction of a trained DNN
• Basic introduction to Python and Theano
• Implementation of DNN, auto-encoder, CNN, RBM with examples:
  handwritten digit- and object-recognition, as well as image restauration /

To the top of the page