After successful completion of the course, students are able to...

• choose suitable methods for a given problem
• employ suitable technologies, software-tools and standards for the solution of a given problem
• understand principles of machine learning

This introductory course teaches the basics of machine learning using computer vision examples. It focuses on a few key areas, on the basis of which the essential laws of machine learning are taught. Students gain a deeper understanding of the methods through independent implementation of the models in the laboratory exercise. The key areas are linear models with fixed basis functions and their kernel extensions as well as optimization methods in supervised and unsupervised learning scenarios. Models with adaptive basis functions and neural networks are introduced. The basic laws that are also dealt with in the exercises are: curse of dimensionality, bias-variance tradeoff and theoretical bounds on the generalization error and the relation to modelcomplexity. We will highlight aspects and challenges of learning from high dimensional data such as images.

In detail, the lecture deals with:

• linear models for regression and classification (Perceptron, linear basis function models, polynomial and radial basis functions, historical overview), applications in computer vision
• pre-processing, feature selection, feature engineering
• neural networks
• error functions and optimization (e.g., pseudo-inverse, gradient descent, newton method)
• model complexity, regularization, model selection, VC dimension 
• kernel methods: duality, sparsity, support vector machine
• principal component analysis (linear encoder) and Hebbian rule
• Bayesian view of the above models, Bayesian regression
• Gaussian processes for regression
• clustering and vector quantization, latent variables (e.g., k-means)
• overview of deep learning models 

An in-depth understanding of the key methodologies is tought by implementing basic classification and regression methods, enhancing them and evaluating them using image data.   Methods are:  

• Lecture
• implementation of methods discussed in the lecture using textual instructions
• carrying out experiments using the implemented methods
• documentation including decription and interpretation of results
• oral feedback during assignment interviews

ECTS Breakdown:

4.5 ECTS = 112.5 hours

30     lecture time

70     2 assignments (including studying machine learning principles, 

       reading documents and literature, 
       implementation of learning algorithms and writing documentation)

2.5    2 interviews (including preparation time)

10     written exam incl. preparation time    

• two assignments
• two assignment interviews
• one written exam

Please register for the course in TISS. After registration you can team up as a group of 3 students in TUWEL.

Knowledge of linear algebra and probability theory