Case Study

Case study: Updating an error analysis spreadsheet when some labels may be wrong.

Case context: A machine learning developer is building a seal-versus-otter image classifier. During error analysis, they suspect that some validation examples have incorrect labels, which could distort the measured error rates. They want to decide whether the amount of mislabeled data is large enough to justify a broader cleanup.

Question: What should the developer add to the error analysis spreadsheet to examine this issue, and what columns should the spreadsheet contain?

Sample answer: The developer should add a 'Wrong label' category so they can measure how much of the validation set is mislabeled. The spreadsheet should have these six columns: File, Seal, Otter, Motion blur, Wrong label, and Notes.

Key points:

  • Add a 'Wrong label' category to track the fraction of mislabeled examples.
  • Use the six columns: File, Seal, Otter, Motion blur, Wrong label, and Notes.

Rubric: A correct response must say that a 'Wrong label' category should be included to track mislabeled examples. It must also list the six columns: File, Seal, Otter, Motion blur, Wrong label, and Notes.

0

1

Updated 2026-08-12

Contributors are:

Who are from:

Tags

Machine Learning

Deep Learning

Supervised Learning

Dive into Deep Learning @ D2L

Data Science

Machine Learning Strategy

Machine Learning Yearning @ DeepLearning.AI