Beschreibung

The course is thought for anyone interested on deep learning and industry applications. It is also a good introduction to the field and specially to the PyTorch deep learning library. In this course the participants will really have hands on and build their own neural networks, not only for typical computer vision tasks, but also for solving more complex problems such as obtaining computer tomography reconstructions.

Learning contents

• Introduction to training neural networks with PyTorch
• Model-based classic approaches, e.g., ISTA.
• Introduction to data-based methods, e.g., LISTA
• Neural Networks for trivial ill-posed inverse problems and fully data-based methods
• Combining model and data-based methods: learned post-processing and learned gradient descent
• Deep Image Prior and mathematical aspects
• Applications on Computed Tomography (CT)

Learning outcomes

At the end of the course the participants will have a good understanding of how neural networks work, and also the mathematical theory behind it. They will also be able to program deep learning approaches themselves using Python and the PyTorch library.

Prior knowledge

Some experience in programming in Python needed.

Technical requirements

Own laptop (for online format a second screen is advantageous), Google Colab (https://colab.research.google.com/)

Further reading

Stanford “Convolutional Neural Networks” course notes: https://cs231n.github.io/

Additional comments

• See flyer KoMSO Academy
• www.uni-bremen.de/techmath/veranstaltungen/2021-komso-academy

Programm/Ablauf