Case Study

Diagnosing a medical imaging classifier with high training error.

Case context: You are training a machine learning model to classify medical images. After the initial training phase, you estimate the algorithm's bias to be 15% and its variance to be 1%. The model struggles to correctly classify the images it was trained on.

Question: Based on these metrics, how would you diagnose the model's current state, and how does it perform on the dev set relative to the training set?

Sample answer: The model's state should be diagnosed as underfitting, characterized by high bias and low variance. Because the variance is only 1%, the model's error on the dev set will be barely higher than its 15% training error.

Key points:

  • The estimated bias of 15% and variance of 1% indicates high bias and low variance.
  • The dev set error is barely higher than the training error.
  • The algorithm is underfitting the data.

Rubric: The response must state that the model is underfitting (or has high bias/low variance) and note that the dev set error is only slightly higher than the training error.

0

1

Updated 2026-06-13

Contributors are:

Who are from:

Tags

Machine Learning

Deep Learning

Machine Learning Strategy

Supervised Learning

Dive into Deep Learning @ D2L

Data Science

Machine Learning Yearning @ DeepLearning.AI