Illustration of the Neural Network Learning Process during Training

This thesis enhances the understanding of Neural Network (NN) trainings by
investigation of the learning process especially focusing on the dependence
of the NN output on the input space for given tasks. For this purpose, the
NN function is decomposed into a Taylor expansion. The Taylor coefficients
serve as a metric to illustrate the influence of input space features on the
output at each step of the training. Both, the arithmetic mean values of the
Taylor coefficients and their dependence on each point of the input space are
investigated, giving new insights into the decision taking of NNs.