A utility inspector has one labeled photo of a new insulator crack and many unlabeled tower photos. The team wants the model to flag that crack after seeing only that single labeled photo. Which learning setup matches this request?
Select an answer to reveal the explanation.
Short Explanation
One photo on the bulletin board is the whole lesson for that crack. You want one-shot learning, which tries to recognize the new crack from that single labeled photo. Few-shot wants several photos, and zero-shot wants a description instead of the photo.
Full Explanation
One-shot learning tries to recognize a new class from a single labeled example. The inspector has one labeled photo of the new crack and wants flags on other tower photos after that lone example, which is the one-shot setup. The unlabeled tower photos may be the pictures to score later, but they do not teach the new crack, because none of them carries that label. Zero-shot learning would use a written description and would not need the labeled photo, so it does not match a request built on that image. Few-shot learning that waits for several labeled photographs of this crack demands more labels than the single photo the team has. Standard supervised training that holds the flag until a large labeled collection exists postpones the task the team is asking to do now. Exam caveat: one-shot still depends on a useful representation learned from other classes, but the new class itself contributes only one labeled photo. Operational check: confirm the training record for the new crack contains exactly one labeled image and that the run is not waiting on a larger labeled set of that crack.