Perceptron

Splashery stretches a drumhead over this Splash's outline, twelve inches across, and finds the ways it can ring. Its lowest is 340.1 Hz, 46 cents flat of F4. The next fifteen are its overtones, and each one draws its own figure: sand gathers on the lines that stay still.

Tap it to try an example. Hands-on: tap an input lamp to switch it; the output updates.

A perceptron is the simplest artificial neuron, described by Frank Rosenblatt in 1958. It takes a few inputs, multiplies each by a weight (how much that input counts), and adds them up. If the sum passes a threshold, it fires: its answer is 1, otherwise 0. It learns by nudging its weights each time it gets an answer wrong.

Tap it: inputs X1 and X3 light up and send pulses along wires as thick as their weights. The sum fills the gauge but stays under the threshold line, so the lamp flashes red. The two live wires thicken, the pulses go again, the gauge passes the line and the lamp lights gold. Pick the flat poster or a 3D model in the Splash tab. This repeats a scripted learning example; it does not keep training on your taps. With ✋ Hands-on on, tap an input lamp to switch it on or off: the sum, the gauge and the output lamp show the perceptron's answer at once, with its learned weights.

The weighted sum and threshold example are consistent with a perceptron. The weight change is a prepared learning example, and tapping does not train a new model.

1. Correct

   A perceptron sums weighted inputs and applies a threshold.¹

   Inputs 1, 0, 1 give sums 0.35 before and 1.35 after, with threshold 1.

2. Simplified

   The weight change illustrates a learning update.¹

   Adding 0.5 times each active input gives the stored second weight set, but the update is not calculated from a live training example.

3. Correct

   Rosenblatt described the perceptron in 1958.¹

   The About text now dates the published description, instead of asserting a precise invention year.

1. 1The perceptron (1958), Frank Rosenblatt, Psychological Review

Simplified

Three fixed inputs and two stored weight sets repeat on each tap. There is no training dataset, learned bias, or ongoing optimizer.

Made in Splashery. Its code is MIT.

This link opens it in the Playground. To put it on your own page, add the element and one script tag. Every option.