Computer Lab Pattern Recognition

Content

The Computer Lab Pattern Recognition is designed to provide students with prior theoretical knowledge of Machine Learning with a practical introduction to the field of Machine Learning and in particular Deep Learning. There will first be an introduction to the Python programming language followed by an introduction to the Deep Learning libraries PyTorch and Tensorflow. This will be followed by practice in the concrete application of these Deep Learning libraries using specific examples from the areas of image processing and language processing. The lab consists of 7 units, of which at least 6 must be passed. The tasks will take place on a Jupyterhub. Every 2 weeks a unit will be made available to be worked on and then handed in. In detail, the following topics will be worked on in the 7 units of the Computer Lab:

• Interactive introduction to Python basics using Jupyter notebooks, Basics of data processing, preparation and visualization.
• Using single-layer machine learning models to solve a two-class problem: support vector machines (based on libsvm) vs. a neural network. Partitioning and use of datasets, application of appropriate metrics for evaluation, use of high-level machine learning libraries such as SciKit-Learn.
• Use of deep neural networks to solve a multi-class classification problem, familiarization with recognized academic datasets such as MNIST and CIFAR-10, introduction to the use of deep learning libraries PyTorch and Tensorflow, use and adaptation of pre-trained models
• Use of convolutional networks to solve more challenging image processing problems such as semantic segmentation and depth estimation, use of regularization methods in training
• Use of manifold cost functions to optimize neural networks, implementation of generative models such as Generative Adversarial Networks (GANs)
• Use of recurrent neural networks to solve problems based on time series data, application of concepts for anomaly detection
• Use of recurrent neural networks for speech processing on the example of noise reduction, analysis of neural networks with respect to their complexity (FLOPs, number of parameters)

Lecturer: Prof. Dr.-Ing. Tim Fingscheidt

Assistant: Marvin Klingner

Lab (PATREC Lab) (Module-No.: TBD):
 
Contact hours (SWS): 4h =  5 LPs
Times: individually
Location: online
Language: German / English

Registration and Procedures

A kick-off event will take place at the beginning of each semester. This will take place in SoSe 2023 on 19.4.2023, 16:45-17:30 in SN 22.1 lecture hall. A registration will then take place until 20.4.2023, 23:59 in the stud.IP event. A subsequent registration is not possible. All further information will be published in stud.IP and at the kick-off.