Learn Before
Case Study

Explain why a voice-command model looks strong on one split but weak on another.

Case context: A voice-command recognition system achieves strong results on the training set and on a small validation set drawn from the same source as the training data. However, it performs much worse on the main development set.

Question: What problem best explains this pattern, and why does it happen?

Sample answer: The most likely issue is a data mismatch between the training data and the development data. The training set and the small validation set are similar, so the model does well there, but the main development set comes from a different distribution. Because the training data is not a good match for the development data, performance drops when the model is evaluated on the development set.

Key points:

  • The correct diagnosis is data mismatch
  • The model does well on the training set and the small validation set, but poorly on the development set
  • The reason is that the training distribution does not match the development distribution

Rubric: The student must identify data mismatch and explain that the training data is a poor fit for the development data distribution.

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