URL study guide
https://tue.osiris-student.nl/onderwijscatalogus/extern/cursus?cursuscode=2MMA80&collegejaar=2025&taal=enOmschrijving
Neural networks have proved to be a versatile and powerful tool in many applications. In this course we develop a mathematical understanding of why that is, by making use of the frameworks of Approximation Theory and the theory of Optimization.We start by introducing the mathematical structure of neural networks and presenting their approximation properties. We compare them to classical approximation classes, such as polynomials and Fourier series, and give arguments why neural networks have better performance in some cases.
In practice, neural networks are trained on data, and the properties and choices made in this training have a strong influence on the final quality of approximation. We analyze some of the commonly used optimization methods, such as gradient descent and stochastic gradient descent, and show how and when they perform well.
On the practical side we introduce the student to the implementation of an end-to-end deep learning solution of a challenging engineering problem using the PyTorch deep learning framework. For this part we expect the student to be familiar with Python.