Case Study

Adding a New Error Category During Review

Case context: You are evaluating a speech recognition system for customer-service calls. While sorting errors, you first tag problems as "low volume" and "overlapping speakers." Later, you notice many failures are actually caused by a new class: heavy background music.

Question: According to the iterative process used in error analysis, what should you do when you identify the "background music" category, and what about the examples you already reviewed?

Sample answer: Add "background music" as a new error category. Then revisit the earlier mistakes you already examined, because some of them may belong in this new group instead of the categories you assigned at first.

Key points:

  • New categories can be added as they appear.
  • Error analysis is iterative, not one-pass.
  • Earlier examples may need to be reclassified after a new pattern is found.

Rubric: Full credit if the learner says to add the new category and to recheck the previously reviewed examples for reclassification.

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