During a machine-learning project at NTNU, I made educational animations that I am quite proud of.
Each point in each frame describes how confident the neural network is that a point is red. Each frame is one learning iteration. Eventually, the network learns the spiral pattern.
Each layer can be thought of as a linear transformation of the points. This animation shows the points transforming from one layer to the next as learning progresses. After the final transformation, a linear projection gives each point a number between 0 and 1, describing the network’s confidence that it is red.