Learn Before
A Human-Comparison Question for Error Review
Question: When you inspect examples the model labeled incorrectly, what human-comparison question can help you identify new error types and possible fixes?
Sample answer: Ask whether a person could label the example correctly, and if so, how.
Key points:
- Consider whether a human would get the example right.
- If so, think about the cues a human would use to do so.
Rubric: The answer must identify the question of whether a human could correctly label the mistaken example and, if helpful, how that person would do it.
0
1
Tags
Machine Learning
Deep Learning
Supervised Learning
Dive into Deep Learning @ D2L
Data Science
Machine Learning Strategy
Machine Learning Yearning @ DeepLearning.AI
Related
What often happens when you inspect misclassified cases during error analysis?
You must finish defining every error category before reviewing any mislabeled examples.
After you discover that many mistakes come from night-time, motion-blurred photos, you should add a new _____ to the error analysis spreadsheet.
Match each review step to its main role in uncovering new error categories.
Order the steps for expanding error categories during review of mistakes.
When examining a misclassified satellite photo, which question is most useful for uncovering new error categories and fixes?
Reviewing a sample of mistakes can uncover a recurring error type that was not included in the original error categories.
Finding New Error Categories
Match each term from an error-review framework to its description.
Put the error-analysis steps in order after you notice a new pattern in reviewed examples.
Why Error Categories Should Keep Evolving
Spotting a New Error Category During Review
A Human-Comparison Question for Error Review