Scribd
Upload
20 views

Final Unit 2 Questions.

Uploaded by

akhbre22
Copyright
© © All Rights Reserved
We take content rights seriously. If you suspect this is your content, claim it here.
Available Formats
Download as XLSX, PDF, TXT or read online on Scribd
20 views

Final Unit 2 Questions.

Uploaded by

akhbre22
Copyright
© © All Rights Reserved
We take content rights seriously. If you suspect this is your content, claim it here.
Available Formats
Download as XLSX, PDF, TXT or read online on Scribd
You are on page 1/ 5

unit 2 section c

subunit
1 1 Explain how data augmentation techniques can improve the performance of a CNN like Alex Net wh
1 2 You are using a pre-trained Alex Net model on ImageNet to classify images of flowers from a new da
1 3 Given the LeNet-5 architecture originally designed for MNIST (grayscale images of size 28×28, descr
2 4 Given an input image size of 64×64×3 and a CNN architecture with two convolutional layers, followe
2 5 Given a CNN with an input of size 32×32×3, a first convolutional layer with 8 filters of size 3×3, stride
2 6 A CNN uses an input size of 64×64×1 and has two consecutive convolutional layers. The first layer ha
3 7 Explain what a dilated convolutional layer is and describe how it differs from a standard convolution
3 8 Given an input feature map of size 32×32 and a dilated convolutional layer with a filter size of 3×3, a
3 9 How does increasing the dilation rate in a dilated convolutional layer affect the computational comp
4 10 What are the key differences between L1 (Lasso) and L2 (Ridge) regularization techniques? Provide e
4 11 How is data augmentation different from regularization techniques? Explain how each approach hel
4 12 What is the dropout regularization technique in neural networks? Describe its primary function

subunit section D
1 1 What does a higher stride value in CNN typically result in?
1 2 What is the role of filters in CNN?
2 3 What type of pooling retains the maximum value from a region?
2 4 What is the effect of zero-padding on the feature map size after convolution?
3 5 Which CNN variant is used to increase spatial resolution during image generation tasks?
4 6 How can CNNs be used in unsupervised learning?
4 7 What is the role of random weights in CNN feature extraction?
4 8 Which unsupervised learning technique can be applied to CNN features for clustering images?
mance of a CNN like Alex Net when trained on a relatively small dataset. Provide specific examples of augmentation methods (e.g., rotation
mages of flowers from a new dataset with 5 classes. Describe the steps you would take to apply transfer learning for this task. Include det
ale images of size 28×28, describe how you would modify it to train on the CIFAR-10 dataset (coloured images of size 32×32×3. Include ch
wo convolutional layers, followed by a max pooling layer, explain how to compute the output size after each layer. Assume the first convol
r with 8 filters of size 3×3, stride 1, and 'same' padding, followed by a ReLU activation, and then a max pooling layer with pool size 2×2 and
lutional layers. The first layer has 32 filters of size 3×3 with stride 1 and 'same' padding. The second layer has 64 filters of size 5×5 with stri
ers from a standard convolutional layer in a neural network. Why might dilated convolutions be useful in tasks such as semantic segmenta
l layer with a filter size of 3×3, a dilation rate of 2, and a stride of 1 with 'same' padding, calculate the size of the output feature map. Show
r affect the computational complexity during the forward pass? If a layer has a dilation rate of 1, and another has a dilation rate of 4 with a
larization techniques? Provide examples of when each technique is preferred
Explain how each approach helps to improve the performance of a machine learning model
escribe its primary function

e generation tasks?

res for clustering images?


ntation methods (e.g., rotation, flipping, scaling) and discuss their impact on the network's ability to generalize.
rning for this task. Include details on which layers you would freeze, how you would modify the network’s output layer, and how you wou
es of size 32×32×3. Include changes to the input layer, any necessary adjustments to convolutional layers, and how you would adapt the f
layer. Assume the first convolutional layer has a 3×3 filter and stride 1 with 'same' padding, and the second convolutional layer has a 5×5
ng layer with pool size 2×2 and stride 2, calculate the total number of parameters in the first convolutional layer. Explain each step in your
s 64 filters of size 5×5 with stride 2 and 'valid' padding. Calculate the output size after each layer.
ks such as semantic segmentation?
the output feature map. Show all your steps.
r has a dilation rate of 4 with all other parameters remaining the same, compare their computational requirements
utput layer, and how you would retrain the model to achieve optimal performance.
nd how you would adapt the fully connected layers for the 10-class output. Provide a rationale for each modification.
convolutional layer has a 5×5 filter, stride 2, and 'valid' padding. The pooling layer has a pool size of 2×2 with stride 2. Provide all layer out
ayer. Explain each step in your calculation.
h stride 2. Provide all layer output sizes on the diagram?

You might also like