Lecture Deep Learning

The lecture will take place Fridays from 8:15 am to 9:45 am. The subsequent Exercise will be held between 10:15 am and 11:45 am. Both will be held in room BIN 1.B-01.

This lecture provides a mathematical understanding on how artificial neural networks and, particularly, deep learning works. In the exercise, we will use a deep learning framework, PyTorch, to employ the theoretical knowledge in practical deep learning examples.

The topics of the lecture will include:

• Two-layer networks for regression and classification
• (Stochastic) gradient descent and error backpropagation
• Convolutional networks and deep architectures
• Generative Adversarial Networks (GANs)
• Recurrent network architectures
• Open-set classification networks
• Adversarial attacks on deep networks

Due to the Corona situation, the lecture will be held both in person and online access is provided through Zoom. Prior to the lecture, slides will be uploaded to OLAT. There, you can also find the link to the Zoom meeting. More details on the lecture can also be found in the VVZ under MINF4568 and DINF1131.

When attending the lecture, it is advisable to download the lecture slides beforehand and have them on an annotatable medium (printed on paper, opened in Acrobat Reader, opened in a tablet computer, ...) such that annotations can be added to the slides.

The exercise will pick up the theoretical contents of the lecture and students will implement the contained mathematics to solve practical examples. Implementations will be carried out in the Python programming language. In the first lectures, neural networks and their learning algorithms will be implemented by hand using Python and NumPy. Later, deeper and more complicated network architectures will be implemented using the PyTorch framework. 

The examination is different for Master and PhD students. 

For Master students (MINF4568), there will be an open-book written examination (50%) and a practical programming task in Python (50%), both taking place on Friday, 18.06.2021, between 08:00 - 10:00 am. Online examination supervision (proctoring) is possible and is agreed upon booking the module. Grades will be provided in quarters between 1-6.

PhD students agree on a short project, possibly from their own field, four weeks after starting the lecture at the latest. At the end of the semester, there will be one or more joint online meetings where each doctoral student discusses their project. Each presentation will include 20 minutes of talk followed by 10 minutes of questions from the audience. The outcome of the project is a pass or fail decision.