This has been a great read so far with incredible diagrams
Here are some favorites:
This diagram shows how weights can parse out patterns in different parts of the image. These weights detect vertical lines and can the activation map below. Note these weights are machine learned! So some networks will create a vertical line detector and some will prefer other detectors.
This is a great diagram showing a dog using a Q Table to value assign values to actions as it navigates as an agent in an environment. When it receives the reward for the “Kick” action at 0 steps away, the Q table updates the 10 reward! This reward can propagate to previous steps in a discounted fashion (notice the 8 reward for “Forward” 8 steps away.
Buy the book Artificial Intelligence: A Guide for Thinking Humans (here)