Deep Learning Lab

Course content

The Deep Learning Lab is aiming to impart knowledge to the students in the fields of machine learning and pattern recognition by practical application of corresponding methods. Students learn to implement and configure classification algorithms, such as linear discriminant functions, support vector machines, and neural networks. Modern concepts and approaches, especially deep learning are also part of the experiments. To motivate subsequent self-study only free-to-use datasets as well as the freely available programming language Python and open-source software will be used. For the computational complex training algorithms students are provided access to powerful centralized GPU (Graphical Processing Unit) hardware. The Deep Learning Lab is divided in three parts: •

1. First, the students work themselves through an introduction to the Python programming language and all required libraries for the later experiments to obtain some basic knowledge. •
2. Second, the students will work with certain machine learning methods which are introduced in the Pattern Recognition lecture. •
3. Third, - in the so-called Machine Learning Challenge - students are required to use their obtained knowledge in order to develop a machine learning system in a competition with the other participating groups.

Therefore, the students will be provided with real data which might stem from real-world/industry applications. To support the ability to work in a team the exercises and the Machine Learning Challenge will be conducted in groups of 2-3 students. The maximum amount of participants is limited to 30 students. If there are more registrations than places available, we will apply a random selection. We recommend to have attended either the lecture Pattern Recognition, or a comparable lecture as a basis for this lab. The results of the first and second praxis phases will be reviewed in a colloquium with the supervising assistants. The systems of the Deep Learning Challenge are presented in a closing event to the other groups, and possibly to representatives of the companies that provided data for the challenge.

• Quality measures in pattern recognition
• Linear discriminant functions, single layer perceptron
• Support vector machines (SVMs)
• Neural networks (NNs)
• Methods for training deep neural networks