Fundamentals of testing AI-based systems · Question 48 of 60

For an image classification model there are many unlabeled test images but no expected results. Which approach fits?

1 answer is correct. Hard

1. A Simply rating the unlabeled images as passed, since no failure can be determined without an expectation
2. B A proxy oracle, for example a model trained on labeled data that assesses the unlabeled tests
3. C Skipping the test until all images have been fully labeled by experts
4. D Using the model under test itself as the oracle and assuming its outputs are correct