CNN Transfer Learning with CNNs 3

Question 1

Which popular pretrained CNN architecture introduced depthwise separable convolutions, making it efficient for transfer learning on mobile devices?

Accepted Answer

MobileNet

Answer

VGG-16

Answer

AlexNet

Answer

ResNet-50

Question 2

When the target dataset is large and very different from the source dataset, the recommended transfer learning approach is:

Accepted Answer

Fine-tune the entire network or train from scratch

Answer

Freeze all pretrained layers

Answer

Fine-tune only the last layer

Answer

Use nearest-neighbor classification on pretrained features

Question 3

What is the purpose of using a lower learning rate for earlier (pretrained) layers compared to later layers during fine-tuning?

Accepted Answer

To preserve general low-level features while adapting high-level ones

Answer

To speed up convergence of the whole network

Answer

To reduce memory usage during backpropagation

Answer

To prevent overfitting the validation set

Question 4

In transfer learning, 'bottleneck features' refer to:

Accepted Answer

Compressed features stored in bottleneck files for fast retrieval

Answer

Features at the narrowest convolutional layer

Answer

Activations from the penultimate layer of a pretrained model

Answer

Features that cause gradient bottlenecks

Question 5

ResNet's skip connections primarily help transfer learning by:

Accepted Answer

Preventing vanishing gradients and enabling very deep pretrained models

Answer

Reducing the number of trainable parameters

Answer

Skipping layers during inference for speed

Answer

Connecting source and target domain layers

Question 6

Which scenario LEAST benefits from transfer learning?

Accepted Answer

General object detection with 10 million diverse images

Answer

Medical imaging with 500 labeled examples

Answer

Satellite imagery classification with 1,000 samples

Answer

Rare species identification with 200 photos

Question 7

What is domain adaptation in the context of CNN transfer learning?

Accepted Answer

Techniques to reduce the distribution gap between source and target domains

Answer

Changing the CNN architecture to match the new task

Answer

Adapting the learning rate schedule during fine-tuning

Answer

Converting image formats to match training data