In this seminar we will discuss the most recent advances and successes of Deep Learning. Therefore, we will provide a brief introduction into Deep Learning itself and then further discuss the most recent techniques and methods presented by the leading research groups of the field. Each participant will review a corresponding publication and summarize its contents in a 45 minute presentation. As a guideline see the book “Dive into Deep Learning” (https://d2l.ai/), Chapters 7 and 17.

The students are supposed to form groups of three, which probably leads to 5-6 groups. (a total of 20 participants are registered) Each group should work on a scientific article and present and discuss it within a maximum of 90 minutes.

1. Nina, Francesco, Ivan (BERT; June 18)
2. Jan-Hendrik, Florian, Alexander (GANs; June 25)
3. Muhammad (GNN; Sep 17)
4. Sinan, Sinan, Riza (CapsNets; Sep 17)

1. Bidirectional Encoder Representations from Transformers (BERT)
   • “Dive into Deep Learning”: Chapter 14 (https://d2l.ai)
   • BERT
2. Generative Adversarial Networks (GAN's)
   • OriginalPaper(2014)
   • Tutorial (2016)
   • “Dive into Deep Learning”, chapter 16
3. Graph Neural Networks (GNN)
   • Benchmarking Graph Neural Networks
   • Note the surveys

• Surveys
  • A State-of-the-Art Survey on Deep Learning Theory and Architectures
  • Recent Advances in Deep Learning: An Overview

You need to login in order to view the literature.