This little network is trying to solve XOR — a classic puzzle a single neuron can't crack. Hit Train and watch the connections adjust, the predictions sharpen, and the error fall as it figures things out.
2 inputs → 4 hidden neurons → 1 output. Blue lines push toward 1, red lines push toward 0. Thicker = stronger.
Steer the training and see how each choice changes the outcome.
How big a step it takes each update. Too small = slow; too big = unstable.
“Loss” measures how wrong the network is. Learning is working when this line trends down toward zero.
The goal: output ≈ 1 when inputs differ, ≈ 0 when they match.
| Input A | Input B | Target | Prediction | Right? |
|---|
Inputs flow through the weighted connections. Each neuron sums what it receives and “fires” more or less strongly.
We compare the output to the right answer. The gap is the loss — our score for how wrong we were.
The error is sent backward to nudge every weight in the direction that reduces it. Repeat thousands of times.