Watch a tiny network teach itself the XOR problem, one gradient descent step at a time.
Bigger steps learn faster but can overshoot. Smaller steps are steadier but slower.
Loss measures how wrong the network is. Gradient descent pushes it downward.
| Input x₁ | Input x₂ | Target XOR | Network Guess | Error |
|---|---|---|---|---|
| 0 | 0 | 0 | — | — |
| 0 | 1 | 1 | — | — |
| 1 | 0 | 1 | — | — |
| 1 | 1 | 0 | — | — |
Each circle holds a number between 0 and 1. Brighter color means a stronger activation.
Connections carry multipliers. The network adjusts these during training to find the right pattern.
XOR can't be solved with a straight line. The hidden layer combines inputs in nonlinear ways.
After each guess, the network measures the error and nudges every weight to reduce it.