Skip to content
Free public beta until December 4, 2026: every course is open, no card needed. After that, the first chapter of each course stays free and full access is a one-time $39 ($19 for up to 100 founding members chosen for helpful feedback). Your progress is kept. How the beta works

label noise

The share of examples carrying the wrong label.

Label noise is the fraction of mislabelled examples; it caps what a model can learn and makes a test score a correct model as wrong.

Examples

  • If a third of the windows labelled "tap" are really a hand held still, the model learns that some stillness is a tap, and more training cannot fix it.

  • Labelling a video afterwards can mark both halves of a tap split across two windows as "tap", adding label noise.

See also

label class balance leakage
Read the full section →