Decoder. plain-English AI glossary

Weights and biases

● Core

The two kinds of knobs a network learns: weights scale each input, biases nudge the result.

Think of it like

A recipe where weights are how much of each ingredient and the bias is the pinch of salt added no matter what.

Example

Training a model means slowly turning billions of these weights and biases until the outputs stop being wrong.

How it actually works

Each neuron multiplies its inputs by weights, sums them, and adds a bias, then applies an activation. Weights decide what matters; the bias shifts the threshold for firing. Every “parameter” you hear about is one of these numbers, and learning is just adjusting them.

For product teams

“Number of parameters” is essentially a count of these knobs — a rough proxy for a model’s raw capacity.

For engineers

The learnable affine parameters W and b in y = Wx + b, updated by gradient descent.

Related

Read anything AI without the jargon

Look up any term in plain English, or save terms as you read with the free Chrome extension.

Open DecoderAdd to Chrome