During the convolution operation in a Convolutional Neural Network (CNN), what does the term "stride" specifically refer to?
Select an answer to reveal the explanation.
Short Explanation
Think of stride like taking steps down a sidewalk. If your stride is one, you step on every single concrete square. If your stride is two, you skip every other square. In a CNN, the filter slides across the image pixels. A stride of 1 means it moves over one pixel at a time. A stride of 2 means it jumps two pixels at a time, which actually shrinks the size of the output feature map. It's a handy way to reduce spatial dimensions without needing a separate pooling layer. So, Option B is your winner!
Full Explanation
In a Convolutional Neural Network (CNN), the stride is a hyperparameter that controls how the filter (or kernel) moves across the input data (such as a 2D image). When performing convolution, the filter is placed on the input, element-wise multiplication is performed, and the results are summed to produce a single value in the output feature map. The filter then shifts by a set number of pixels defined by the stride. If the stride is set to 1, the filter slides one pixel at a time. If the stride is set to 2, the filter shifts by two pixels at a time, skipping intermediate positions. Increasing the stride reduces the overlap between receptive fields and decreases the spatial dimensions of the output feature map. Let's look at the other options: Option A defines the number of filters, which determines the depth (number of channels) of the output feature map, not the stride. Option C describes the kernel size (e.g., 3x3 or 5x5). Option D refers to the output spatial dimensions, which are influenced by the stride but are not the stride itself. For the exam, remember that stride dictates the movement step size of the kernel across the input.