Backpropagation Explained Intuitively — 3Blue1Brown

A LITTLE CONTEXT BEFORE YOU WATCH

Your viewing guide

This lesson follows the learning problem inside a neural network: how should its connections change when an answer is wrong? Backpropagation organizes the influence of those connections so training can adjust them.

Why choose this one?

Choose it after you understand the basic layers-and-weights picture. The valuable shift is from saying that a network should improve to asking which internal changes would help. Following that chain of influence is more useful than memorizing a term without a model of it.

Who it’s for

For learners who already recognize a neural network and want an intuitive bridge into training. If the diagrams feel too abstract, return to the handwritten-digit introduction first. A visual explanation supports, but does not replace, later mathematical practice.

What to notice and try

• Keep the desired answer and the current answer separate as you follow the example.
• Notice how a change nearer the output can depend on activity earlier in the network.
• Revisit the creator’s written lesson when you want to work through the explanation more slowly.

Creator or broadcaster’s background notes ↗

Related watches

• Neural Networks Explained Visually — 3Blue1Brown
• Vectors Explained Visually — 3Blue1Brown

Viewing notes by TubeShowcase · Reviewed 2026-10-29. Films and broadcasts remain the work of their credited creators.