Essay

Why Label Cleanup Can Be Acceptable in Some Settings

Question: In a concise analytical response, explain why correcting uncertain or noisy labels may be acceptable in one context but not in another.

Sample answer: If the goal is to improve a deployed system or guide product iteration, adjusting labels after reviewing mistakes can be reasonable because the main objective is better practical performance. However, if the result will be reported in a research paper or used as an unbiased estimate of model accuracy, the same adjustment becomes a problem because it can distort the measurement. In short, acceptability depends on whether the intended use can tolerate bias or requires a neutral evaluation.

Key points:

  • Reasonable for product or system improvement
  • Problematic for research reporting
  • Problematic for an unbiased accuracy estimate
  • Intended use determines whether the bias matters

Rubric: A strong response states the development-use exception, identifies the research and unbiased-evaluation cases as problematic, and explains that acceptability depends on the goal.

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