IT lexicon AI & ML Gradient descent

Gradient descent

AI & ML På svenska → Updated: 2026-05-23

The optimisation algorithm that trains almost every ML model — take small steps in the direction that reduces error most.

Analogy: you're on a mountain in fog and want to get down. Feel the slope, walk downhill, repeat. Variants: SGD, Adam, AdamW — all built on the same idea with different smart tricks.

← Back to the lexicon