In the two part lecture “Actuarial calculations on the basis of Deep Neural Networks (DNN),” we would like to show how to use deep neural networks for actuarial calculations.
The first video shows that deep neuronal networks are universal function approximators. It quickly demonstrates what happens inside a DNN when a value is calculated and what happens between the different layers to achieve a decent approximation of a mathematical function. The goal is to give the viewer an insight as to how deep neuronal networks can approximate mathematical functions.
The second video gives a brief, hands-on example in which I build, train and evaluate a deep neural network to approximate the calculation of a tariff premium. We go through the steps of loading and preparing the data, building a DNN, training it with the data and evaluating the model. Additionally, a pretrained and much better model is loaded and evaluated to show, what might be possible. All the source code is executed in real time.