CNN Classic CNN Architectures 2 — Questions and Answers
Question 1: What was the primary innovation of AlexNet that enabled its 2012 ImageNet victory?
- Use of average pooling instead of max pooling
- Training on GPUs with ReLU activations and dropout (Correct answer)
- Introduction of batch normalization after every layer
- Use of 1x1 convolutions to reduce parameters
Correct answer: Training on GPUs with ReLU activations and dropout
AlexNet used GPU training, ReLU activations (avoiding vanishing gradients), and dropout regularization to win ILSVRC 2012 by a large margin.
Question 2: How many convolutional layers does the original VGGNet (VGG-16) contain?
- 8
- 13 (Correct answer)
- 16
- 19
Correct answer: 13
VGG-16 has 13 convolutional layers and 3 fully connected layers, totaling 16 weight layers.
Question 3: What is the key structural unit introduced by GoogLeNet (Inception v1)?
- Residual block
- Inception module (Correct answer)
- Dense block
- Depthwise separable convolution
Correct answer: Inception module
GoogLeNet introduced the Inception module, which applies 1x1, 3x3, and 5x5 convolutions in parallel and concatenates their outputs.
Question 4: In ResNet, what problem do skip connections primarily solve?
- Overfitting on small datasets
- Vanishing and exploding gradients in very deep networks (Correct answer)
- Excessive memory usage during inference
- Slow convergence due to poor weight initialization
Correct answer: Vanishing and exploding gradients in very deep networks
Skip connections allow gradients to flow directly through the network, mitigating vanishing/exploding gradient problems in very deep architectures.
Question 5: Which architecture first demonstrated that network depth beyond 20 layers could be effectively trained?
- AlexNet
- VGGNet
- ResNet (Correct answer)
- SqueezeNet
Correct answer: ResNet
ResNet (2015) successfully trained networks with 50, 101, and even 152 layers using residual connections, proving extreme depth was achievable.
Question 6: What filter size does VGGNet use exclusively in its convolutional layers?
- 1x1
- 3x3 (Correct answer)
- 5x5
- 7x7
Correct answer: 3x3
VGGNet uses only 3x3 convolutional filters, showing that stacking small filters achieves the same receptive field as larger filters with fewer parameters.
Question 7: What is the depth of the winning ResNet model in the ILSVRC 2015 competition?
- 50 layers
- 101 layers
- 152 layers (Correct answer)
- 200 layers
Correct answer: 152 layers
The ILSVRC 2015 winning ResNet used 152 layers, achieving a top-5 error rate of 3.57%.
CNN Classic CNN Architectures 2
What was the primary innovation of AlexNet that enabled its 2012 ImageNet victory?