Machine Learning 101 • Interactive

Watch a tiny neural network learn XOR.

Press Train and see gradient descent gently adjust connection weights until the network learns a classic pattern: output 1 when the two inputs are different, and 0 when they match.

Current loss

—lower is better

The network

Circles are neurons. Lines are weighted connections: blue pushes a signal up, pink pushes it down. Thicker lines mean stronger weights.

2 inputs3 hidden neurons1 output

Training complete: the loss is low!

1. Forward passThe two input numbers flow through the network to make a prediction.

2. Measure errorLoss compares the prediction with the correct XOR answer.

3. Backprop + updateGradient descent nudges each weight in the direction that reduces loss.

Training controls

Try different learning rates. Too small learns slowly; too large can wobble.

Learning rate0.50

carefulbold

Epoch

Accuracy

Live loss curve

Each point is one training snapshot. A healthy learner slopes downward.

XOR truth table

The bars show the network’s current output for each example.

Why hidden neurons? A single straight-line rule cannot solve XOR. The hidden layer lets the network combine simple bends into a smarter boundary.

✓After training, matching inputs should produce outputs near 0.
✓Different inputs should produce outputs near 1.