Case Study

Evaluating the feasibility of artificial data synthesis for a specialized validation set

Case context: A machine learning team has a development set with very specific audio noise characteristics. They have clean speech data and want to use synthesis to generate training data. However, they are unsure if their proposed synthesis pipeline is worth the effort.

Question: According to the principles of dev-set matching via artificial data synthesis, what core criteria must their synthesis process satisfy to justify its implementation?

Sample answer: The synthesis process must occur under circumstances where it allows the team to create a huge dataset, and this synthesized dataset must reasonably match the specific conditions of the dev set. If the synthesized data is too small or does not match the dev set distribution, the effort is not justified.

Key points:

  • The synthesis must produce a huge dataset
  • The synthesized data must reasonably match the dev set
  • It must address the gap between available training data and the dev set

Rubric: Look for the student to identify that the synthesis must allow the creation of a huge dataset and that this dataset must reasonably match the dev set distribution.

0

1

Updated 2026-05-26

Contributors are:

Who are from:

Tags

Machine Learning

Deep Learning

Supervised Learning

Dive into Deep Learning @ D2L

Data Science

Machine Learning Strategy

Related