AlgoPlus//machine learning / gradient-descent
Read the theory

Gradient Descent

Roll downhill on the loss surface — the learning rate sets the step size.

w
Loss
Iter
Learning rate
Legend
Current weight
Gradient / path
Minimum
AI Tutor Workspace
In a nutshell
Gradient descent is how most models learn. It measures the loss — a single number for how wrong the model currently is — then nudges the model's numbers in the direction that lowers the loss fastest, and repeats. The size of each nudge is set by the learning rate: too small and training crawls, too large and it overshoots the bottom and bounces instead of settling.
Ready
Press play to begin the cinematic walkthrough.
You're blindfolded on a hill and want the bottom. Feel which way is steepest down, take a step, repeat. The step size (learning rate) decides whether you glide in or bounce around.
Key terms
Go deeper in the lesson
Read the full theory, intuition & complexity for Gradient Descent.