Spiking neural networks are notoriously hard to train because of their complex dynamics and sparse spiking signals. However, in part due to these properties, spiking neurons possess high computa- tional power and high theoretical energy efficiency. This thesis introduces an online, supervised, and gradient-based learning algorithm for spiking neural networks. It is shown how gradients of temporal signals that influence spiking neurons can be calculated online as an eligibility trace. The trace rep- resents the temporal gradient as a single scalar value and is recursively updated at each consecutive iteration. Moreover, the learning method uses approximate error signals to simplify their calculation and make the error calculation compatible with online learning. In several experiments, it is shown that the algorithm can solve spatial credit assignment problems with short-term temporal dependencies in deep spiking neural networks. Potential approaches for improving the algorithm’s performance on long-term temporal credit assignment problems are also discussed. Besides the research on spiking neural networks, this thesis includes an in-depth literature study on the topics of neuromorphic computing and deep learning, as well as extensive evaluations of several learning algorithms for spiking neural networks

End-to-end trained Convolutional Neural Networks have led to a breakthrough in optical flow estimation. The most recent advances focus on improving the optical flow estimation by improving the architecture and setting a new benchmark on the publicly available MPI-Sintel dataset. Instead, in this article, we investigate how deep neural networks estimate optical flow. By obtaining an understanding of how these networks function, more can be said about the behavior of these networks in unexpected scenarios and how the architecture and training data can be improved to obtain a better performance. For our investigation, we use a filter identification method that has played a major role in uncovering the motion filters present in animal brains in neuropsychological research. The method shows that the filters in deep neural networks are sensitive to a variety of motion patterns. Not only do we find translation filters, as demonstrated in animal brains, but thanks to the easier measurements in artificial neural networks, we even unveil dilation, rotation and occlusion filters. Furthermore, we find similarities in the refinement part of the network and the perceptual filling-in process which occurs in the mammal primary visual cortex.