Highway Networks and Skip Connections Quiz

Answers and explanations appear as you take the quiz.

1. A very deep network struggles with optimization. What can a skip path provide?
   • A guarantee of perfect gradients at every depth
   • A more direct signal and gradient route that can ease training
   • A proof that more layers always improve accuracy
   • A replacement for the loss function
2. What does a basic residual block compute?
   • output = F(x) multiplied by x
   • output = x divided by F(x)
   • output = the maximum of F(x) and x
   • output = F(x) plus x, where x is the input shortcut
3. What inspired the gating mechanism in highway networks?
   • LSTM gates from recurrent networks
   • Convolutional pooling
   • Batch normalization
   • Dropout
4. In a highway network, what does the transform gate T(x) control?
   • The learning rate of the optimizer
   • The size of the output image
   • How much of the input is transformed versus carried straight through
   • The number of layers in the network
5. Which modern architecture relies on residual connections around its sublayers?
   • K-means clustering
   • The decision tree
   • Linear regression
   • The transformer
6. How do U-Net skip connections specifically help segmentation?
   • They delete the encoder entirely
   • They pass fine spatial detail from the encoder to the decoder
   • They convert the image to text
   • They reduce the number of output classes to one
7. In y = F(x)T(x) + x(1-T(x)), what remains when T(x)=0?
   • Only F(x)
   • The sum 2x
   • The carried input x
   • A mandatory zero output
8. For the same coupled-gate highway form, what remains when T(x)=1?
   • The transformed value F(x)
   • Only the original x
   • F(x) plus x
   • An automatically normalized x