Sandbox 02

Build a Brain That Can't Be Fooled

Two sensors, three hidden neurons, one output — and a rule no single neuron can learn alone (sound the alarm when exactly one sensor fires). Watch the hidden layer light up as it trains, and see exactly why deep learning needed more than one layer to begin with.

See this sandbox in its lesson context
Sensor ASensor BAlert?

53%

alert probability

The rule this network is learning: sound the alert when exactly onesensor is triggered — not neither, not both. Try setting both sliders to 0, then both to 1, then one of each. A single neuron can never learn this pattern no matter how its weights are tuned; that’s the entire reason this network needs a hidden layer.

Epoch 0 · 2/4 training examples correct