Case Study

Determine pipeline fault when the cat breed classifier receives a cropped image containing only rocks.

Case context: A Siamese cat detection pipeline fails to detect a cat because the cat detector crops a pile of rocks instead of the cat. The breed classifier receives this cropped image and outputs y=0 (no Siamese cat).

Question: Based on the provided context, diagnose this failure. Identify which component is at fault, and justify your choice using the classifier's output and human classification standards.

Sample answer: The error must be attributed to the cat detector. The breed classifier is blameless because it was given a cropped image of a pile of rocks and correctly classified it as y=0 (not containing a Siamese cat). A human shown the same crop would also predict y=0. Because the breed classifier performed reasonably on the input it was actually given, it is not at fault for the final incorrect pipeline decision; the failure was caused by the cat detector cropping the wrong region.

Key points:

  • The fault is attributed to the cat detector.
  • The breed classifier is blameless because it correctly processed its inputs.
  • An output of y=0 is correct and reasonable for a pile of rocks.
  • A human presented with the same crop would also output y=0.

Rubric: The learner must attribute the fault to the cat detector, state that the breed classifier is blameless because its output of y=0 is correct for the rock input, and note that a human would agree with this classification.

0

1

Updated 2026-05-26

Contributors are:

Who are from:

Tags

Machine Learning

Deep Learning

Supervised Learning

Dive into Deep Learning @ D2L

Data Science

Machine Learning Strategy

Related