Case Study

Diagnose a scenario where actual user uploads differ from the dev/test set

Case context: A team discovers that their actual user uploads (the actual distribution they need to do well on) consist of images that are vastly different from the images they currently have in their dev and test sets.

Question: Based on this discovery, what is the underlying issue with their current evaluation setup, and what specific action should the team take to fix it?

Sample answer: The underlying issue is that their dev/test set distribution is not representative of the actual distribution they need to do well on. Because the actual distribution is different, the team should update their dev/test sets to be more representative of the actual user uploads.

Key points:

  • The actual distribution is different from the dev/test sets.
  • The current dev/test set distribution is not representative.
  • The team must update the dev/test sets to be more representative.

Rubric: The student must identify that the dev/test set distribution is no longer representative of the actual distribution and recommend updating the dev/test sets.

0

1

Updated 2026-05-27

Contributors are:

Who are from:

Tags

Machine Learning

Deep Learning

Machine Learning Strategy

Supervised Learning

Dive into Deep Learning @ D2L

Data Science

Machine Learning Yearning @ DeepLearning.AI

Related