Case Study

Diagnosing primary errors given 10% training, 11% training-dev, and 12% dev rates

Case context: You are evaluating a new image recognition model. Your logs show the following performance: a 10% error rate on the training set, an 11% error rate on the training-dev set, and a 12% error rate on the dev set.

Question: Based on these specific error metrics, what should you diagnose as the primary problem with this algorithm, and what does this imply about its current performance on the training data?

Sample answer: I would diagnose the model as having high avoidable bias. This implies that the algorithm is currently doing poorly on the training set, meaning it is underfitting or failing to capture the underlying patterns in the training data effectively.

Key points:

  • Diagnose high avoidable bias.
  • Conclude that performance on the training set is poor.
  • Recognize that the model fails to learn the training data adequately.

Rubric: The response must correctly diagnose the issue as 'high avoidable bias' and explicitly state that this means the algorithm is performing poorly on the training set.

0

1

Updated 2026-06-12

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