TensorFlow Developer Certificate Practice Test

1. In TensorFlow, what does the Keras Sequential API primarily allow you to build?

Linear models only

Stacked neural networks layer by layer

Distributed training systems

Custom loss functions

2. Which optimizer is commonly used for training deep neural networks due to its adaptive learning rate?

Stochastic Gradient Descent (SGD)

Adam

RMSprop

Adagrad

3. When using tf.data.Dataset, what is the primary benefit of the .prefetch() method?

Reduces model accuracy

Overlaps data loading with model training

Decreases batch size

Normalizes input features

4. In a CNN, what is the primary purpose of a pooling layer?

Increase feature map dimensions

Reduce spatial dimensions and extract dominant features

Add non-linearity to the network

Normalize input data

5. What does model.fit() return in TensorFlow/Keras?

The trained weights

A History object containing training metrics

The test accuracy

A compiled model

6. Which loss function is typically used for multi-class classification problems?

Binary crossentropy

Mean squared error

Categorical crossentropy

Hinge loss

7. What is the output shape of a Flatten layer when applied to a (batch_size, 28, 28, 3) input?

(batch_size, 28, 28, 3)

(batch_size, 2352)

(batch_size, 84)

(batch_size, 1)

8. In transfer learning, what is typically frozen when fine-tuning a pre-trained model?

The output layer only

Early layers with learned general features

The batch normalization parameters

The loss function

9. What does the model.evaluate() method return?

Training loss only

Loss and metric values on test data

Predictions for new samples

Model architecture summary

10. Which normalization technique scales input features to a range of [0, 1]?

Standardization

Min-max normalization

Batch normalization

Layer normalization

11. In a Recurrent Neural Network (RNN), what is the primary advantage of using LSTM cells over basic RNN cells?

Faster training time

Ability to capture long-term dependencies

Reduced model size

Lower computational cost

12. What is the purpose of dropout regularization in neural networks?

Speed up training

Prevent overfitting by randomly deactivating neurons

Normalize layer outputs

Increase model capacity

13. When saving a TensorFlow model, what does the .save() method preserve?

Architecture only

Weights only

Architecture, weights, and training configuration

Training history only

Submit