ANNs are at the very core of DL. They are versatile, powerful, and scalable, making them ideal for tackling large and highly complex ML tasks. We can classify billions of images, power speech recognition services, and even recommend that hundreds of millions of users watch the best videos, by stacking multiple ANNs together. These multiple stacked ANNs are called Deep Neural Networks (DNNs). Using DNNs, we can build very robust and accurate models for predictive analytics.
The architectures of DNNs can be very different: they are often organized on different layers. The first layer receives the input signals and the last layer produces the output signals. Usually, these networks are identified as Feed-Forward Neural Networks (FFNNs). In this chapter, we will construct an FFNN that classifies an MNIST dataset. Later on, we will see two more implementations of FFNNs (for building very robust and accurate models for predictive analytics...