Formed in the 19th c. from Latin 'gradiēns' (stepping) — the rate at which a surface 'steps' up or down, physical or mathematical.
An inclined part of a road or railway; the degree of a slope; a change in the value of a quantity across a distance.
Formed in English from Latin gradientem, accusative of gradiens, present participle of gradi (to walk, to step, to advance), modelled on quotient and transient. The word was coined in the 19th century to describe the rate of slope — how steeply a surface steps upward or downward per unit of horizontal distance. In mathematics, the gradient generalises this to multi-dimensional functions: it is the vector pointing in the direction
In machine learning, 'gradient descent' is the fundamental optimization algorithm: the model adjusts its parameters by repeatedly 'stepping' in the direction of steepest descent on the error surface. The word 'gradient' — from Latin 'stepping' — is literally what the algorithm does: it takes steps downhill. The learning rate controls how large each step is. Too large and you overshoot; too small and learning is glacially slow.