Learn Before
Identify which component caused the mistake when the second-stage classifier saw the wrong crop.
Question: A document-processing system first detects a receipt and then sends the cropped region to a yes/no classifier that decides whether the crop contains a receipt. The classifier outputs y=0 for a crop that actually contains only a blank tabletop. Why is the detection stage, rather than the classifier, credited with the mistake?
Sample answer: The detector is responsible because the classifier received a crop that did not contain the intended object. Given a blank tabletop, predicting y=0 is a reasonable and correct output, so the classifier behaved appropriately on the input it was given.
Key points:
- The classifier’s output y=0 matches the actual content of the crop.
- A person shown the same crop would also likely say there is no receipt.
- The upstream detector made the real error by selecting the wrong region.
Rubric: The answer should explain that the classifier is not at fault because it labeled its input appropriately, while the detector caused the problem by passing an incorrect crop to the classifier.
0
1
Tags
Machine Learning
Deep Learning
Supervised Learning
Dive into Deep Learning @ D2L
Data Science
Machine Learning Strategy
Machine Learning Yearning @ DeepLearning.AI
Related
Which stage should be blamed if a two-step wildlife classifier crops the wrong image region and the second model correctly says the crop does not contain the target species?
True or False: If a dog-breed classifier labels a tightly cropped dog photo correctly, it can still be blamed for an error caused by the crop itself.
If a document pipeline crops the wrong area from a photo of a shipping label and the text model correctly says that crop does not contain a tracking number, the error belongs to the _____ step.
When the package scanner crops a photo of a delivery truck’s tire instead of the shipping label, and the label classifier correctly outputs y=0, which part should be blamed for the pipeline error?
If an upstream cropper selects a blank patch of pavement from a photo of a delivery truck, a downstream truck classifier is not to blame when it outputs y=0.
When the face classifier labels a badly cropped image as y=0, the pipeline mistake is attributed to the _____.
Match each part of the processing chain to its role in a crop-based classification error case.
Order the steps for tracing a package-sorting error to the correct component.
Why is the last-stage classifier considered blameless when a prediction pipeline gives the wrong final result?
In a two-stage pet classifier example, the mistaken crop passed to the breed model shows a stack of cardboard boxes.
A correct negative prediction on a clean input should not be blamed on the model.
Match each result in the document-processing example to what it tells you about the source of the error.
Order the steps in a document pipeline when a bad crop causes the final error.
Attributing pipeline error when a detector sends the wrong crop to a species classifier
Find the defective stage when a cropper sends the wrong image patch to a species classifier.
Identify which component caused the mistake when the second-stage classifier saw the wrong crop.