Statistische und maschinelle Lernverfahren - Heinz Nixdorf Institut (HNI)

Statistical and Machine Learning

Brief description

Example of a two-layer, fully connected neural network

The course Statistical and Machine Learning Methods provides an insight into the components and algorithms of statistical and machine learning methods. Various approaches are presented on how characteristics can be learnt from data, either supervised or unsupervised, and how unknown patterns can be recognised. The techniques presented can be applied to a variety of classification and regression problems, whether for one-dimensional signals (e.g. language), two-dimensional (e.g. images) or symbolic data (e.g. texts, documents).

Lecture contents

Introduction to classification methods, Bayesian and other decision rules
Optimisation methods: Gradient descent, algorithmic differentiation, optimisation with boundary condition
Linear classifiers: perceptron, support vector machines
Deep neural networks
Dimension reduction methods (PCA, LDA)
Unsupervised learning: mixture distributions, clustering methods

Learning outcomes & professional competences

After attending the course, students will be able to

select a suitable solution method for a given classification or regression problem
Apply methods of supervised and unsupervised learning to new problems and critically evaluate the results of the learning process
Have a basic understanding of machine learning methods
Can use program libraries to implement classifiers (e.g. neural networks, support vector machines) and write their own programs
can make a sensible choice for the dimension of the feature vector and the complexity of the classifier for a given set of training data

The students

Have acquired skills in Python that they can also use outside this application domain
Have an understanding of the principle of parsimony and can transfer it to other problems
Can analyse a given classification or regression problem, synthesise a solution and then evaluate it on test data
Can transfer the knowledge and skills gained in this course to other disciplines
Can analyse more extensive tasks together in a group, break them down into subtasks and work on them in a solution-oriented manner
Can assess the performance and limitations of machine learning methods

Methodical realisation

Lectures with predominantly blackboard use, occasional slide presentation
Classroom exercises with exercise sheets and demonstrations on the computer
practical exercises with Python, in which students independently generate training and test data, work out solutions and implement and test learning procedures or classifiers and analyse results.