The MNIST dataset does not contain numbers on the edges of images. Hence, neither network assigns relevant values to the pixels located in that region. Both networks are much better at classifying numbers correctly if we draw them closer to the center of the designated area. This shows that neural networks can only be as powerful as the data that is used to train them. If the data used for training is very different than what we are trying to predict, the network will most likely produce disappointing results. In this chapter, we will cover the basics of neural networks and how to set up a deep learning programming environment. We will also explore the common components of a neural network and its essential operations. We will conclude this chapter by exploring a trained neural network created using TensorFlow.
This chapter is...