Summary
In this chapter, we explored deep learning from the inside. We saw that building a CNN is now easy with TensorFlow 2.x, but peeking inside the way it "thinks" gives critical insight.
We first built a CNN with many layers. The level of abstraction of a CNN increases through each layer. Reducing the number of dimensions per layer makes patterns appear. A neural network can be described as a process that goes from chaos to meaning.
After building the CNN, we wrote a program that can read the "mental" images of the layers. The output of each layer shows how the network is creating patterns and structures. Since we humans often think using mental images, the output images of the CNN help us understand how a machine learns.
Finally, we used a Google Colaboratory server to visualize the measurements of the CNN's learning process with TensorBoard running on top of TensorFlow 2.x. Measuring the accuracy of the training process of a CNN...
