Module S3.13 - Image Processing III

Coordinator

Prof. Dr. Michael Habeck

Lecturers

Prof. Dr. Michael Habeck, Prof. Dr. Cord Spreckelsen, Dr. Holger Babovsky

Schedule

• Lectures/Exercises: every Thursday 09.00-12.00 (SR Kollegiengasse 10), starting on Oct 16th.
• !!Please enrol to the module in FRIEDOLIN!!

Contents

• Recap regression and curve fitting, recap linear algebra
• Supervised and unsupervised machine learning
• Shallow and deep neural networks
• Basic approximation theory (Why does it work?, Universal approximation theorem)
• Loss functions (classification, regression)
• Training deep neural networks by stochastic gradient descent (SGD) (“Backprop”)
• Practical experience with training a simple network using Pytorch, e.g. fitting a simple one-dimensional function (1D input / 1D output)
• Regularization, overfitting, training and test data
• Convolutional neural networks, image classification, medical applications
• Networks for specific image processing tasks (e.g. U-Net for segmentation, denoising)
• Transformers, Attention, Vision Transformers
• Generative AI (diffusion models, flow models)

Why does it work?