Summary

In this chapter, we introduced NNs in detail, and we mentioned their success vis-à-vis other competing algorithms. NNs are comprised of interconnected units, where the weights of the connections characterize the strength of the communication between different units. We discussed different network architectures, how an NN can have many layers, and why inner (hidden) layers are important. We explained how information flows from the input to the output by passing from one layer to the next, based on weights and the activation function. Finally, we showed how to train NNs – that is, how to adjust their weights using GD and BP.

In the following chapter, we’ll continue discussing deep NNs. We’ll explain in particular the meaning of deep in deep learning, and that it not only refers to the number of hidden layers in a network but to the quality of the learning of the network. For this purpose, we’ll show how NNs learn to recognize features and compile...