Learning Deep Learning: Theory and Practice of Neural Networks, Computer Vision, NLP, and Transformers using TensorFlow

Chapter 4

Fully Connected Networks Applied to Multiclass Classification

In the first three chapters, we used our neural network to solve simple problems that set a foundation for learning deep learning (DL). We reviewed the basic workings of a neuron, how multiple neurons can be connected, and how to devise a suitable learning algorithm. Combining this knowledge, we built a network that can act as an XOR gate—something that arguably can be done in a simpler way.

In this chapter, we finally get to the point where we build a network that does something nontrivial. We show how to build a network that can take an image of a handwritten digit as input, identify which one of the ten digits 0 through 9 the image represents, and present this information ...