問題文
A team replaces a rule-based image sorter with a trained image classifier. The old sorter used conditions that developers had written by hand, while the new classifier was fitted to a labeled photo set. An auditor asks the team to show why any single photo was routed the way it was. Why is that request harder to satisfy now?
選択肢
- The behavior of the classifier comes from the listed instructions that the design team recorded in the code file, so the path of each decision has no buried tag.
- The behavior of the classifier comes from the compiled bytecode that the build stage optimized in the target binary, so the path of each decision has no source maps.
- The behavior of the classifier comes from the statistical patterns that the training stage discovered in the photo set, so the path of each decision has no readable listing.
- The behavior of the classifier comes from the fixed thresholds that the vendor manual documented in the release notice, so the path of each decision has no learned part.