Advanced Deep Learning

In this continuation module to "Fundamentals of Deep Learning", some further modeling concepts and learning techniques from the deep learning area are to be learned and understood, in particular the transformer architecture is to be understood and first simple examples of generative models are to be learned.

RNN

Students know the most important types of Recurrent Neural Networks (RNN) and their use (as classifier, sequence-to-sequence, encoder/decoder). They understand the weaknesses of 'simple' RNNs and how these can be mitigated with the help of long-term memory in GRUs or LSTMs. You will be able to implement some example applications independently.

Attention, Transformer

Students understand the concept of the attention mechanism and how this is implemented in the Transformer architecture (e.g. for language translation).

Foundation Models, Self-Supervised Learning

Students have already learned about the most important characteristics of foundation models in "Fundamentals of Deep Learning". Here they will gain a deeper understanding of how foundation models are trained. Particular attention will be paid to the frequently used "Self-Supervised Learning" and "Contrastive Learning" in this context.

Generative Models (Introduction)

The students know some simple examples of generative models and how they can be trained and applied. Variational Autoencoders (VAE) or Generative Adversarial Networks (GAN), for example, serve as a good introduction. They can implement these in a suitable framework and understand the difficulties that can arise during training.