Banach Center – Oberwolfach Graduate Seminar: Mathematics of Deep Learning

Teaching material for 'Mathematics of Deep Learning Workshop' (https://www.mfo.de/occasion/1947a)

Overview

Talks

1. Gitta Kutyniok

   • Introduction to Neural Networks
   • Approximation Theory and Expressivity I
   • Deep Learning meets Inverse Problems
   • Deep Learning meets Parametric Partial Differential Equation
   • Interpretability

2. Philipp Grohs

   • Mathematical Foundations of ML
   • Approximation Theory
   • Deep Neural Networks for PDEs
   • Generalization for Deep Learning

3. Julius Berner

   • NN Training in the Overparametrized Setting

Notebooks

Google Colaboratory

1. Philipp Grohs

   • Linear Regression with LASSO: Linear_Regression_with_LASSO

2. Julius Berner

   • A simple approach to the Fashion-MNIST dataset: Fashion_MNIST

   • Framework for constructing deep neural networks to efficiently approximate various functions: NN_Approximation

   • Deep learning based method for solving high-dimensional Kolmogorov PDEs: DL_Kolmogorov

Local Environment

1. Install the Python development environment
2. Create a virtual environment
3. Install requirements.txt

(see https://www.tensorflow.org/install/pip)