IT lexicon AI & ML Backpropagation

Backpropagation

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

The algorithm that trains neural networks — runs error backwards through the network and adjusts every weight.

Steps: run input forward → get output → compare to right answer → compute gradient for each weight (chain rule from calculus) → nudge weights toward the right direction. Repeat billions of times.

← Back to the lexicon