Learn Before
Essay

Explain how error analysis by parts helps prioritize improvements in a machine learning system.

Question: In a multi-stage machine learning system, what is the main reason to perform error analysis by parts? Also, describe a quick informal way to do it using a human-performance reference.

Sample answer: The main purpose of error analysis by parts is to decide which part of the system deserves the next round of engineering effort. Instead of treating all mistakes as equal, you break the pipeline into components and estimate how much each one contributes to the overall failures. A practical informal approach is to compare each component to a human benchmark for the same subtask, then compare the full system to that benchmark as well. The parts that are farthest from human performance are usually the most promising targets for improvement.

Key points:

  • The goal is to rank components by how much improvement they are likely to yield.
  • Analyze mistakes component by component rather than only at the end-to-end level.
  • Use human performance as a rough reference point for each subtask.
  • Compare both individual modules and the complete system against that reference.

Rubric: The response should explain that the point of the method is prioritization of fixes and should describe the informal human-benchmark comparison for both subcomponents and the overall system.

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