Enable JavaScript to grade the quiz and save your result. You can still read the explanations below.
Which example uses a self-supervised signal?
Expected answer. Mask a span and predict the text that was masked
The target comes from the text itself. No annotator supplies the expected answer.
Review section 2.1When evaluating next month’s performance, what should happen to case threads?
Expected answer. Keep each thread in one split and preserve the time cutoff
A grouped time split keeps familiar threads out of the test set.
Review section 2.2Policy A flags 104 requests, 40 of them urgent. There are 50 urgent cases in total. What is its precision?
Expected answer. 40/104
Precision is measured among flagged cases. The fraction 40/50 measures recall.
Review section 2.3Training loss keeps falling while validation loss rises. Which decision leaves the test independent?
Expected answer. Choose a stage on validation loss, then evaluate once on test
Validation selects the stage. The test remains reserved for the final measurement.
Review section 2.4What allows a stack of dense layers to represent non-affine boundaries?
Expected answer. Nonlinear activation functions
Composing affine maps still yields an affine map. A nonlinear activation changes that property.
Review section 2.5