問題文
A team measures the accuracy of its model, finds it above the agreed figure, and plans no further model testing. Which set of risks does the syllabus list for a model beyond the failure to reach a performance measure, together with their mitigations?
選択肢
- A model can render slowly, it can hold too loosely to its input buffer, and it can fail on a broken font, so cache and paging and style all need their own tests.
- A model can cost heavily, it can hold too firmly to its idle nodes, and it can strain on a busy day, so budget and sizing and quota all need their own tests.
- A model can behave unfairly, it can hold too tightly to its training set, and it can fail on a crafted input, so bias and fitting and attack all need their own tests.
- A model can end quickly, it can hold too long to its signed terms, and it can clash on a shared host, so renewal and audit and occupancy all need their own tests.