AlexNet Architecture Quiz

What is the filter/kernel size for the first convolutional layer from the original AlexNet model?

• 96

• (11, 11, 3)

• (11, 11)

• (96, 11, 11)

How many classes were in the ImageNet dataset on which the original AlexNet architecture was trained?

• 10

• 4096

• 1000

• 227

How many filters does the second convolutional layer (conv2) use?

• 96

• 256

• 512

• 128

From the options below, pick all the layers on which we can set pretrained weights.

• pool1

• conv3

• conv1

• fc8

• flatten

What does layer freezing mean?

• Drop the temperature of the layer below 0

• Stop the layer from updating its weights

• Stop the layer from updating its biases

• Stop the layer from updating its weights and biases

What function can we use to change the values of the weights and biases from a model without having to go through the training process?

• update_weights

• set_weights

• set_weights_biases

• update_weights_biases

How can we stop layers from updating their weights and biases values during the training process?

• set the "trainable" attribute to False

• set the "trainable" attribute to True

• set the "layer" attribute to True

• set the "layer" attribute to False

We want to use a pretrained model on a dataset that has a different number of classes than the dataset that the model was originally trained on. Which layer must we to retrain?

• fully connected output layer

• average pooling layer

• first convolutional layer

• all convolutional layers

• all pooling layers

Which type of layer usually has the largest number of parameters in a model?

• pooling

• convolutional

• fully connected

• flatten

Out of the models used in the workshop notebook, which one is the deepest?

• VGG16

• ResNet50

• Pretraine AlexNet

• Original AlexNet