Neural networks and deep-learning evolve to a technology embedded into many different aspects of our every-day routine.

Often, these techniques are viewed as black boxes without insights to their inner workings. Visualization techniques provide both developers and Users important insights into the functionality of neural networks. This can lead to further optimization of deep learning approaches, easier explanations for the functionality of those networks, usability even for non-experts and broader acceptance of such technologies.

In this seminar, we are going to investigate and analyze visualization techniques in the field of deep learning. Our question will be: How can training and analyzing neural networks be made more transparent?

The seminar is held as a block event at the end of the semester, while the topics will be assigned at the beginning of the semester. Each participant prepares a report, and gives a 20 minute presentation during the block event. Furthermore, all participants will be involved in the reviewing process of their peer's papers. The papers are improved based on the feedback resulting from the review process.

All deadlines are strict deadlines, and no extensions will be granted.

Dates for plenary meetings can be discussed at the first meeting to suit individual schedules.

To successfully complete the seminar the following requirements have to be fulfilled:

• Written paper on a selected topic, German or English (English preferred)
• 20 minutes presentation with questions and answers at the end, German or English
• Active participation in peer reviews and discussions
• Meet the schedule