Foundation in Machine Learning

Data is being produced and available to us in ever growing amounts, in industrial processes, but also in everyday life. We can use these data to understand patterns, optimize processes and make predictions. What constitutes a valuable prediction depends on the data and the application. The large amounts of data available further require tools to summarize, group and visualize them in appropriate ways. Methods that enable to tackle these tasks with computer assistance belong to the field of machine learning. In this module, we introduce fundamental machine learning methods and applications. For a number of standard algorithms and methods we will learn their mathematical foundations, how to implement and apply them and how to evaluate their use.

Basic understanding and procedure Students will understand the structure of supervised and unsupervised learning tasks. They can further identify or appropriately structure datasets that can be used to predict continuous and categorical variables. Students are familiar with the modelling process and relevant aspects of model selection (pre-processing, train-test-validate, feature selection, bias-variance decomposition, overfitting, hyperparameter tuning). They can evaluate regression and classification models appropriately by using different metrics and critically examine prediction results. Finally, students know possible industrial applications of basic supervised and unsupervised machine learning methods .

Theory understanding and methods Supervised Learning Students know a series of standard methods for regression and classification (linear regression, lasso, ridge regression, kernel ridge regression, CART, kNN, MLP, logistic regression, SVM, Naive Bayes, LDA, etc.) and can use them properly. They are familiar with the underlying mathematical models and algorithms and can explain how they work. In particular, students know for the different models about approaches to assess and adjust model complexity.

Theory understanding and methods Unsupervised Learning Students know a series of standard methods of unsupervised learning (PCA, SVD, NMF, ICA, hierarchical clustering, DBSCAN, kMeans, k-Medoids, t-SNE, UMAP, Isolation Forests, LOF, GMM) and can describe their mathematical foundations, as well as the hyperparameters they introduce. Students know how applications of unsupervised learning methods can be examined and evaluated quantitatively and graphically (dissimilarity analysis, elbow plot, silhouette plot, ... ).

Application and practical understanding Students know the properties, advantages and disadvantages of the above supervised and unsupervised machine learning methods and can use them purposefully by employing dedicated libraries (e.g. scikit-learn). They are able to systematically find suitable models through diagnostic visualizations and appropriate metrics. Students can discuss the application of machine learning methods in a technically correct manner using examples from their accumulated experience. They know about challenges that can arise from the structure of datasets (e.g. class imbalance, outliers) and about approaches how to cope with such circumstances. Finally, students know about the possible combined use of unsupervised and supervised learning methods.