Case Study

How should a team choose beta when equal total weight is only a starting point?

Case context: A team optimizes an objective whose first sum contains 5,000 mobile images and whose second sum contains 200,000 auxiliary internet images. It begins with beta = 1/40 but wants to consider alternative weights.

Question: Diagnose what beta = 1/40 currently does and decide how the team should evaluate other beta values using the supplied evidence.

Sample answer: The current beta value gives the mobile-image collection and internet-image collection equal total weight despite their different sizes. If the team wants to consider a different balance, it should try other beta values and tune the parameter on the dev set.

Key points:

  • Beta = 1/40 equalizes the collections' total weights.
  • The second sum contains the auxiliary internet images.
  • Alternative beta values are permitted.
  • The dev set should guide tuning among those values.

Rubric: The response should explain the effect of beta = 1/40 and recommend dev-set tuning for alternative beta values. It should keep beta associated with the auxiliary internet-image data.

0

1

Updated 2026-07-20

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