This section is designed to introduce the core components of NNs and deep learning. For those who already are familiar with NNs, it may feel like a condensed overview of the topic, but feel free to jump to the next section if you are here only for the practical tips about Keras.

An NN, or ANN to avoid any confusion, is a powerful method that can approximate any sort of function, linear or not. If you don't know anything about ANNs, here you will get the basics: the main components, and how the training takes place. You might learn which are the hyperparameters and algorithms to choose while building a network. We will discuss matters such as the following ones:

• How many nodes should I use in each layer?
• Which activation functions should I use?
• How is data transformation likely to alter the results?
• Which error measure should I adopt?
• Which training algorithm (training strategy) should I pick?
• What kind of network should I have?

Deciding upon a network and its architecture is usually a serious consideration, as the preceding list may have demonstrated, and requires many decisions to be made. Moreover, as generic modeling tasks are frequently tagged as artistic quests, due to how much creativity they demand, so also is NN modeling.

Creativity is certainly related to the creation and development of NNs. The idea that inspired NNs has been floating around for quite a long time. As the name suggests, it came from neuroscience and was based on a basic understanding of how neurons work.