Virtual Labs

1. What is Transfer Learning?

a: Training a CNN model from scratch every time Explanation

Explanation

b: Using a pretrained model for a new related task Explanation

Explanation

c: Using only test data for training Explanation

Explanation

d: Removing convolution layers permanently Explanation

Explanation

2. Why is Transfer Learning useful for small datasets?

a: It increases the number of classes automatically Explanation

Explanation

b: It removes the need for training Explanation

Explanation

c: It reduces overfitting by using pretrained features Explanation

Explanation

d: It converts images into text Explanation

Explanation

3. Which large dataset is commonly used to pretrain deep CNN models?

a: MNIST Explanation

Explanation

b: ImageNet Explanation

Explanation

c: Oxford Flowers Explanation

Explanation

d: IMDB Reviews Explanation

Explanation

4. Which of the following is a lightweight pretrained CNN model?

a: MobileNet Explanation

Explanation

b: VGG19 Explanation

Explanation

c: Inception-ResNet Explanation

Explanation

d: AlexNet Explanation

Explanation

5. In Transfer Learning, “freezing layers” means:

a: Removing layers from the model Explanation

Explanation

b: Preventing weight updates during training Explanation

Explanation

c: Increasing the learning rate of all layers Explanation

Explanation

d: Training only the input layer Explanation

Explanation

6. The early layers of a CNN generally learn:

a: Very complex object-specific patterns Explanation

Explanation

b: Only flower names Explanation

Explanation

c: Basic features like edges and textures Explanation

Explanation

d: Only background noise Explanation

Explanation

7. Fine-tuning in Transfer Learning refers to:

a: Training the entire CNN model from scratch Explanation

Explanation

b: Retraining some pretrained layers on the target dataset Explanation

Explanation

c: Using only validation data for training Explanation

Explanation

d: Converting CNN into an SVM model Explanation

Explanation

8. Transfer learning works best when:

a: Target task has limited training data and a related pretrained model is available Explanation

Explanation

b: Target task always has unlimited data Explanation

Explanation

c: Only text inputs are given for image models Explanation

Explanation

d: No preprocessing is performed Explanation

Explanation

9. Which layer is commonly used to reduce spatial dimensions in CNNs?

a: Pooling layer Explanation

Explanation

b: Dense layer Explanation

Explanation

c: Dropout layer Explanation

Explanation

d: Softmax layer Explanation

Explanation

10. What is a major risk when transferring from an unrelated source dataset?

a: Faster training always occurs Explanation

Explanation

b: Negative Transfer Explanation

Explanation

c: Automatic dataset cleaning Explanation

Explanation

d: Perfect accuracy is guaranteed Explanation

Explanation

Transfer Learning with Deep CNNs