CNN Transfer Learning with CNNs 4

Question 1

When freezing layers during transfer learning in Keras/TensorFlow, which attribute is set to False?

Accepted Answer

layer.trainable

Answer

layer.active

Answer

layer.frozen

Answer

layer.requires_grad

Question 2

Which pretrained model is known for its 'inception modules' that process features at multiple scales simultaneously?

Accepted Answer

GoogLeNet (Inception)

Answer

VGG-19

Answer

DenseNet

Answer

SqueezeNet

Question 3

What problem does data augmentation specifically help address during transfer learning fine-tuning?

Accepted Answer

Overfitting when the target dataset is small

Answer

Catastrophic forgetting of source task

Answer

Slow convergence due to frozen layers

Answer

Gradient explosion in deep layers

Question 4

In PyTorch, how do you freeze all parameters of a pretrained model before adding a new classifier?

Accepted Answer

for param in model.parameters(): param.requires_grad = False

Answer

model.freeze()

Answer

model.eval()

Answer

torch.no_grad()

Question 5

EfficientNet scales CNN depth, width, and resolution using a principle called:

Accepted Answer

Compound scaling

Answer

Progressive resizing

Answer

Layer-wise adaptive rate scaling

Answer

Residual scaling

Question 6

A model pretrained on ImageNet and transferred to a medical X-ray dataset shows poor performance even after fine-tuning. What is the MOST likely cause?

Accepted Answer

Large domain shift between natural photos and X-rays

Answer

ImageNet has too many classes

Answer

The CNN architecture is too shallow

Answer

The learning rate was too low

Question 7

What is 'task-specific fine-tuning' in contrast to 'feature extraction' transfer learning?

Accepted Answer

Updating pretrained weights (some or all) for the new task

Answer

Training only the softmax layer on the new task

Answer

Using task-specific data only for augmentation

Answer

Extracting features specific to the source task