DenseNet and Dense Connectivity Quiz

Answers and explanations appear as you take the quiz.

1. What operation does DenseNet use to combine feature maps from preceding layers?
   • Channel-wise concatenation
   • Element-wise addition
   • Max pooling across layers
   • Element-wise multiplication
2. In DenseNet, what does the 'growth rate' k control?
   • The learning rate schedule
   • The pooling window size
   • The number of new feature maps each layer produces
   • The number of dense blocks
3. How many direct connections exist among L layers in a dense block?
   • L
   • L(L+1)/2
   • 2L
   • L squared
4. What is the main purpose of transition layers in DenseNet?
   • To add residual shortcuts
   • To increase the growth rate
   • To apply dropout only
   • To downsample and reduce feature-map size between dense blocks
5. A key benefit of dense connectivity is that it helps mitigate which training problem?
   • Overfitting to color
   • Exploding batch sizes
   • The vanishing-gradient problem
   • Label noise
6. Compared to ResNet at similar accuracy on ImageNet, DenseNet-BC typically uses:
   • Substantially fewer parameters
   • Far more parameters
   • About the same parameters
   • No convolutions at all
7. A late DenseNet layer needs an edge feature learned near the input. Why can it use that map directly?
   • Only the final layer can see edge features
   • DenseNet resets all maps after each layer
   • Transition layers reconstruct every earlier map from scratch
   • Earlier feature maps remain among its concatenated inputs
8. A dense block begins with m channels and three layers each add k new maps. How many channels emerge before a transition?
   • 3m + k
   • m + 3k
   • m times k cubed
   • Only k, because old maps are discarded