How Models Learn

Gradient

The steepness of the loss along every parameter at once, gathered into one arrow; it points uphill.

The steepness becomes a gradient: an arrow made of one steepness per dial, together pointing straight uphill.
From Gradient descent, step “Two dials, one landscape”

See it in action

Try gradient hands-on in the lesson Gradient descent, part of the AI fundamentals course.

Open the lesson