We now have the knowledge of multiple neural network architectures; it’s time to train them. Before we can train any neural network, we must be sure that the data and model are prepared properly. In most cases, attempting to learn from raw data will yield poor results. Proper normalization is key for the model to more fairly assess the data and prevent optimization issues. Further, once we decide on a neural architecture to use, it is important to consider how the model’s parameters will be initialized. Naive initialization approaches can doom a model’s learning potential from the start. In this chapter, we discuss commonly used normalization and initialization approaches as well as the justifications for using them.