Case Study

Should Extra Labeled Data Be Added to a High-Variance Bird Call Classifier?

Case context: You are building an audio classifier that identifies bird species from short sound clips. The model has high variance, and you have found a large archive of unlabeled field recordings that could be labeled and added to training. At the same time, the compute budget for training and retraining is almost gone.

Question: Using the standard guidance for reducing variance, decide whether adding this new data is a practical next step right now, and explain the reason.

Sample answer: It is not a practical next step right now. Even though there is a large supply of additional data that could be labeled, the team does not have enough compute left to train effectively on it. This remedy only works when both conditions are met: sufficient extra data and enough computational capacity to use it.

Key points:

  • Recognize that the model has high variance.
  • Note that more data is available.
  • Explain that the plan is not practical because compute is insufficient.
  • State that both extra data and adequate compute are required for this approach.

Rubric: The student must say that adding training data is not currently a practical solution. The explanation must make clear that the available data condition is satisfied, but the compute requirement is not satisfied because the training budget is exhausted.

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