Learn Before
Why the data source does not change a consistent label mapping
Question: A plant-disease classifier is trained with photos of the same leaves, some taken in a greenhouse and others taken outdoors. Why does the function f(x) not need to know which setting produced the photo in order to decide whether this auxiliary data source is consistent with the target task?
Sample answer: Because the label depends on the visual content of the leaf, not on whether the photo was taken in a greenhouse or outdoors. The same input-to-label mapping applies in both settings, so f(x) can predict y from x without using the source of x.
Key points:
- The mapping from input x to label y is the same for both settings.
- The source of the photo is not needed to determine the label.
Rubric: The answer should explain that f(x) can map x to y without knowing whether the photo came from the greenhouse or outdoors because the underlying label relationship is unchanged across the two sources.
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
What makes an auxiliary data source consistent with a target task?
A document classifier must know whether a news article was written on a laptop or a phone in order to predict its topic correctly.
A helper dataset is compatible with the target task when the same _____ can assign labels in both places.
Match each term to its role when deciding whether a side dataset is a consistent auxiliary source.
Order the steps for checking whether a supplemental data source matches the main prediction task.
What is the main practical drawback of adding a consistent auxiliary data source during training?
A training source that follows the same labeling rule as the main task is usually low risk to include and can sometimes improve performance.
A Wildlife Classifier Does Not Need Image Provenance
Match each idea about auxiliary data to its meaning.
Order the steps for deciding whether to add a second data source to training.
Consistency Between Two Labeled Data Sources
Evaluating Consistent Auxiliary Data for a Damage Detector
Why the data source does not change a consistent label mapping