ImageNet Classification with Deep CNN

Who is the primary author of the paper "ImageNet Classification with Deep Convolutional Neural Networks"

• Ilya Sutskever

• Alex Krizhevsky

• Yann LeCun

• Geoffrey Hinton

Which activation function did the AlexNet architecture use?

•  $f(x)=\tanh(x)$  
  

•  $f(x)=(1+e−x)^{-1}$  

•  $f(x)=\max(0,x)$  

Choose the max pooling configuration used in the AlexNet architecture.

• 
• 
• 

AlexNet uses a technique for reducing complex co-adaptation of neurons. Which of the following do they correspond to?

• 50% of all activations of the network were set to zero and also did not participate in backpropagation

• Local response normalization

• Combining prediction of different models

• Dropout of 0.5

What type of neural network does AlextNet use?

• Feedforward Neural Network

• Perceptron

• Convolutional Neural Network

• Long Short Term Memory

What approaches of data augmentation did AlexNet opt?

• Adding random crops of images, their rotations and flips in training

• Light conditioning

• Test time augmentation by prediction on corner and center patches of image

• PCA color augmentation

On what did the AlextNet train on?

• CPU

• 1 GPU

• 2 GPU

• 3 GPU

Which of the following statements are true based on the graph?

• Implies the reason why the network uses saturating activations

• The dotted lines corresponds to tanh and solid line to ReLU

• Use of ReLU increased training speed of the network by six times

• The dotted lines corresponds to ReLU and solid line to tanh

How many layers did AlexNet have?

• 8 Convolutional layers

• 5 Convolutional and 3 fully connected

• 8 Fully connected layers

• 3 Convolutional layers and 5 fully connected

Which of these do you consider the most crucial to the success of the architecture.

• Down sampling images to fixed resolution of 256 x 256

• ReLU non linearity

• GPU cross parallelization

• Normalization layer

• Overlapping pooling