The fourth session will address the question "How is a neural network trained on data? " in depth. The goal of this session is to understand the learning process of a neural network. We will therefore have a closer look on numerical optimization and how to make "learning goals" quantifiable. Different objective functions are investigated and distance metrics are discussed. Further, we will introduce you to the method of backpropagation.

To train a network, we have to SEE what is going on. Together we will plot the loss during training and go through different scenarios. What do we do when the loss is no longer getting smaller? How to get out of a potential local minima?