250047 VU Applied machine learning (2022W)

Machine learning plays an essential role almost everywhere in technology and research these days. The creation of data-driven models, which can be used efficiently as proxy and/or prediction models for new data, requires "know-how" in the mathematical basics as well as experience in dealing with the appropriate software packages.
The VU aims to "reconcile" theoretical and application-relevant aspects. In addition to the particularly practically relevant basics such as data analysis, model selection, validation, over- and underfitting, "feature selection", we will discuss "classical methods" such as non-linear model reduction (PCA, MDS, kernel methods, etc.), classification (logistic regression, random forests , SVMs, etc), regression (kernel rigde regression, lasso, etc), clustering, ensemble learning, as well as the basics of the current Deep Neural Networks (DNN), especially new advanced methodologies like autoencoders/-decoders, convolutional Neural Networks and " Physics-Informed Neural Networks (PINN)".

Using practical exercises, the handling of this is taught "hands-on" via Python, scikit-learn and keras/tensor flow, where we use the no-setup environment Kaggle with free GPU access. The VU will also deal with current research topics and applications in physics and materials research.

Structure:
The VU "Applied Machine Learning" combines (i) a "theory part" where mathematical and numerical basics of machine learning are presented, (ii) practical exercises that accompany the first part, and (iii) a "group work" where a small application problem in groups of about 2-6 people is considered and presented.