Item analysis
Question statistics
| Question | n | Difficulty p | Discrimination r_pb | Verdict |
|---|---|---|---|---|
| How does the skip connection in a ResNet unit help mitigate the vanishing gradient problem? p=1.00 > 0.85 — almost everybody answers correctly | 6 | 1.00 | n/a | too easy |
| How does the constant error carousel in an LSTM help preserve gradients over long sequences? | 6 | 0.83 | 0.29 | OK |
| What does gradient clipping do to handle exploding gradients in RNNs? | 6 | 0.83 | 0.29 | OK |
| What is the main advantage of weight reuse in convolutional neural networks? | 6 | 0.83 | 0.29 | OK |
| What is the primary purpose of label smoothing as introduced in Inception v3? | 6 | 0.83 | 0.29 | OK |
Rejection rule (from 5 attempts): a question is flagged when its difficulty p falls outside 0.30–0.85 (share answering correctly) or its point-biserial discrimination drops below 0.15 (correlation with the rest of the attempt score). Retiring and regenerating flagged questions runs offline: uv run python -m app.quality … --regenerate.