What Does Kernel Size Do Neural Network

Deciding optimal kernel size for CNN

Convolutional Neural Networks (CNNs) are neural networks that automatically extract useful features (without manual manus-tuning) from information-points similar images to solve some given job like prototype classification or object detection. And at present that you understand their use on your datasets, you lot outset wondering: Autonomously from tuning diverse hyper-parameters of your network, how do I know what is the right kernel size for the network? Let's dig further!

Let the states set some mutual ground rules to stay on the same platform throughout the discussion:

We will be looking primarily at 2D convolutions on images. These concepts also apply for 1D and 3D convolutions, merely may non correlate directly.
A 2nd convolution filter similar 3x3 will always have a third dimension in size. The tertiary dimension is equal to the number of channels of the input image. For example, we apply a 3x3x1 convolution filter on gray-scale images (that has 1 black and white channel) whereas, we apply a 3x3x3 convolution filter on a colored image (with 3 channels, red, blueish and green).
We are bold zero padding for rest of the give-and-take.

In a convolution, a convolution filter slides over all the pixels of the image taking their dot product. Nosotros do this hoping that the linear combination of the pixels weighted by the convolutional filter extracts some kind of feature from the epitome. It is washed keeping these things in mind:

About of the useful features in an image are usually local and it makes sense to take few local pixels at a time to use convolutions.
Most of these useful features may be institute in more than than 1 place in an image. So, it makes sense to slide a single kernel all over the image in the hope of extracting that feature in unlike parts of the image using the aforementioned kernel.
Likewise, an added benefit of using a small kernel instead of a fully connected network is to benefit from weight sharing and reduction in computational costs. To briefly explain this point, since we use the same kernel for different set of pixels in an image, the same weights are shared across these pixel sets as we convolve on them. And every bit the number of weights are less than a fully connected layer, we have bottom weights to back-propagate on.

Now that we have convolution filter size as one of the hyper-parameters to choose from, a choice needs to be made between smaller or larger filter size. Let united states quickly compare both to choose the optimal filter size:

Comparing smaller and larger convolutional kernel sizes theoretically.

Now that we have some idea nearly the extraction using different sizes we will follow this upwards with a convolution case for small (3x3) and large filter sizes (5x5):

Comparison smaller and larger convolutional kernel sizes using a 3x3 and a 5x5 example.

Based on the comparison above, we tin conclude that smaller kernel sizes are and should be a popular option over larger sizes.

Also, you might notice a preference for odd number as kernel size over a 2x2 or 4x4 kernel size. The explanation is as follows:

For an odd-sized filter, all the previous layer pixels would be symmetrically around the output pixel. Without this symmetry, we will take to account for distortions across the layers which happens when using an even sized kernel. Therefore, even sized kernel filters are mostly skipped to promote implementation simplicity. If you think of convolution as an interpolation from the given pixels to a centre pixel, we cannot interpolate to a middle pixel using an even-sized filter.

Therefore, in general, we would like to use smaller odd-sized kernel filters. But, 1x1 is eliminated from the listing candidate optimal filter sizes as the features extracted would be fine grained and local, with no information from the neighboring pixels. Also, it is not really doing whatsoever useful feature extraction!

Hence, 3x3 convolution filters work in general, and is oftentimes the pop pick!