Maschinelles Lernen in der Signalverarbeitung

Module "Vertiefung Mobile Nachrichtensysteme" (ET-12 10 17)

The lectures and tutorials will take place on-site. 

Exercise: Tuesday (every 2nd week) from 9:20am to 10:50am in BAR/I86C/U

Lecture: Friday from 9:20am to 10:50am in BAR/I86C/U

Topics

This lecture is an introduction to statistical machine learning and its applications for modern signal processing problems. The lecture focuses on the design principles of machine learning algorithms.

First we will examine basic methods for regression and classification: linear regression, logistic regression and the k-Nearest-Neighbor algorithm. Using these examples, we will discuss the basic tradeoff between the flexibility of a model and its generalization capability. We will examine properties of learning in high dimensional spaces compared to learning in low dimensional spaces.

Furthermore, we will investigate methods that help to make linear models flexible: polynomials and splines. We will treat wavelets as structured signal representations and discuss the meaning of sparsity in signal representations. This will lead us to compressed sensing and other modern methods for signal denoising, signal reconstruction and signal compression. We will give an overview of key concepts of convex optimization and discuss support vector machines.

The last part of the lecture focuses on neural networks, the backpropagation algorithm, and deep learning.