Build machine-learning models by connecting nodes, then train them and watch every number change.
Examples
Fit a lineLinear regression: learn a slope and an offset that best fit 2-D points.XOR with a neural networkA straight line can’t separate XOR. A small MLP with a hidden layer can.The classic MLP, neuron by neuronEvery neuron and every weight on screen: watch each weight’s chart change as the network learns.Classify handwritten digitsOne Linear layer turns 784 pixels into 10 digit scores.Karate club: a graph neural networkTwo members are labeled; a GNN infers everyone else’s side from the friendships.Why the graph matters: GNN vs MLPSame noisy data, two models: the one that listens to friends wins.A tiny transformerLearns Shakespeare one character at a time, then writes its own.Diffusion from scratchThe idea behind image generators, on 2-D points: learn to remove noise, then turn noise into data.