H13-311_V3.5 · Question #365
In a convolutional neural network, different layers have different functions. Which of the following layers can play the role of dimensionality reduction?
The correct answer is B. Fully connected layer C. Convolutional layer D. Pooling layer. Pooling layers (D) are the most explicit dimensionality reducers - operations like max pooling shrink spatial dimensions (e.g., 2×2 max pool halves height and width). Convolutional layers (C) can reduce dimensions via stride > 1 or "valid" (no) padding, and can also compress…
Question
In a convolutional neural network, different layers have different functions. Which of the following layers can play the role of dimensionality reduction?
Options
- AInput layer
- BFully connected layer
- CConvolutional layer
- DPooling layer
How the community answered
(23 responses)- A22% (5)
- B78% (18)
Explanation
Pooling layers (D) are the most explicit dimensionality reducers - operations like max pooling shrink spatial dimensions (e.g., 2×2 max pool halves height and width). Convolutional layers (C) can reduce dimensions via stride > 1 or "valid" (no) padding, and can also compress channel depth by choosing fewer filters. Fully connected layers (B) perform dimensionality reduction whenever the output size is smaller than the input - mapping, say, 1024 features down to 128 is exactly that. The input layer (A) is wrong because it is purely passive: it receives raw data and passes it forward unchanged, performing no computation or transformation whatsoever.
Memory tip: Ask yourself "does this layer shrink the data?" The input layer is a door - it just lets data in. Everything downstream (conv, pool, FC) can act as a funnel, making the representation smaller.
Topics
Community Discussion
No community discussion yet for this question.