Case Study

Managing new categories during a computer vision error analysis.

Case context: You are building a cat classifier and examining misclassified images. You initially create categories for 'dogs' and 'blurry images'. Halfway through your dev set, you notice several misclassifications are actually lions.

Question: According to the iterative process described in Machine Learning Yearning, how should you handle the discovery of the 'lions' category, and what must you do with the images you already reviewed?

Sample answer: You should add 'lions' as a new error category. Because error analysis is an iterative process, you must then re-examine the misclassified images you have already categorized to see if any of them should be re-assigned to the new 'lions' category.

Key points:

  • Add the newly discovered category.
  • Error analysis is iterative.
  • Re-examine previously categorized examples in light of the new category.

Rubric: Full credit if the learner correctly identifies that the new category should be added and explicitly states that previously examined images must be re-examined.

0

1

Updated 2026-05-27

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