Short Answer

Why some error analysis methods need a human baseline

Question: Briefly explain why certain error analysis methods do not work well when a model contains steps that people cannot reliably do.

Sample answer: Many error analysis methods compare the system against what a person can do. If a component performs a task that people cannot do reliably, then there is no trustworthy human benchmark for that step. Without that reference point, those methods cannot tell whether the component is truly failing or simply operating in a task where human comparison is not meaningful.

Key points:

  • Some error analysis methods depend on a human-performance benchmark.
  • If people cannot do the task well, that benchmark does not exist.
  • Without the benchmark, those methods are not useful for that component.

Rubric: The answer should mention the need for a human benchmark and explain that when humans cannot perform the task well, the benchmark is unavailable, so the related error analysis methods cannot be applied.

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