Meet the Artificial Neuron
See that a neuron just multiplies its inputs by weights, adds them up and adds a bias.
A straight line is fine for house prices, but faces, speech and language need far richer patterns. For those we use neural networks, built from tiny calculators called neurons.
A neuron takes a few inputs (numbers). It multiplies each input by its own weight, which says how much that input matters: a big weight means it matters a lot, a negative weight means it argues against. Then it adds everything up, plus one extra number, the bias, a starting amount that shifts the result up or down. The total is its output.
That is all a neuron does: multiply, add. Its weights and bias are its parameters, and they are everything it "knows". In the simulation you set them by hand. A few lessons from now, the network finds them by itself.
Key takeaways
- A neuron multiplies each input by a weight, adds them up and adds a bias.
- Weights say how much each input matters; negative weights push the answer the other way.
- The bias is a starting amount that makes the neuron easier or harder to "switch on".
- Weights and biases are the parameters: the only things a network learns.
Check your understanding
Go deeper (more technical)
In symbols: output = w₁·x₁ + w₂·x₂ + … + wₙ·xₙ + b. Most neurons then pass this total through a squish function (an "activation function") such as ReLU, which turns negative totals into 0, or sigmoid, which squishes any total into the range 0 to 1. Without that bend, stacking neurons would only ever produce straight lines.
The idea goes back to 1943 (McCulloch and Pitts) and the "perceptron" of 1958. Real brain neurons are far more complex; artificial neurons only borrowed the rough idea of weighted inputs.