Neural Network Convolutional Neural Networks 2 β Questions and Answers
Question 1: What is depthwise separable convolution and what is its main advantage?
- A convolution that processes depth information only
- A factored convolution applied depthwise then pointwise, dramatically reducing parameters and computation (Correct answer)
- A convolution that separates positive and negative activations
- A 3D convolution for volumetric data
Correct answer: A factored convolution applied depthwise then pointwise, dramatically reducing parameters and computation
Depthwise separable convolution applies a single filter per input channel (depthwise) then combines channels with 1Γ1 convolution (pointwise), drastically reducing computation vs standard convolution.
Question 2: What is feature map visualization and why is it used in CNN analysis?
- Visualizing the training loss over time
- Displaying the intermediate activations to understand what features each layer detects (Correct answer)
- Plotting the distribution of filter weights
- Showing the confusion matrix of predictions
Correct answer: Displaying the intermediate activations to understand what features each layer detects
Feature map visualization displays the activations of intermediate layers, revealing what patterns (edges, textures, shapes) the network has learned to detect at each stage.
Question 3: What is transfer learning in the context of CNNs?
- Transferring a model from one GPU to another
- Reusing pretrained CNN weights as a starting point for a new task (Correct answer)
- Converting a CNN model to run on mobile devices
- Sharing weights between encoder and decoder networks
Correct answer: Reusing pretrained CNN weights as a starting point for a new task
Transfer learning uses a CNN pretrained on a large dataset (like ImageNet) as the starting point for a new task, leveraging learned feature representations to reduce training time and data requirements.
Question 4: In CNN-based image classification, what does the global average pooling layer do before the final classifier?
- Applies max pooling across all feature maps simultaneously
- Averages each entire feature map to a single value, reducing spatial dimensions to 1Γ1 (Correct answer)
- Pools gradients globally for stability
- Concatenates all feature maps into a single vector
Correct answer: Averages each entire feature map to a single value, reducing spatial dimensions to 1Γ1
Global average pooling computes the mean of each feature map, converting spatial feature maps of any size into a fixed-length vector suitable for classification.
Question 5: What does data augmentation accomplish during CNN training?
- It increases the model's depth automatically
- It artificially expands the training set by applying random transformations to existing samples (Correct answer)
- It normalizes pixel values to zero mean
- It generates synthetic labels for unlabeled data
Correct answer: It artificially expands the training set by applying random transformations to existing samples
Data augmentation applies random transforms (flips, crops, rotations, color jitter) to training images, artificially increasing dataset diversity and reducing overfitting.
Question 6: What is Grad-CAM and how is it used?
- A gradient-based optimizer for CNNs
- A visualization technique using gradients to highlight image regions important for a prediction (Correct answer)
- A type of convolutional layer with learned pooling
- A camera-based data collection tool for training
Correct answer: A visualization technique using gradients to highlight image regions important for a prediction
Grad-CAM (Gradient-weighted Class Activation Mapping) uses the gradients of the target class flowing into the final conv layer to produce a heatmap highlighting influential image regions.
Neural Network Convolutional Neural Networks 2
What is depthwise separable convolution and what is its main advantage?