Convolutional neural networks are a special form of neural network. In a traditional neural network, the input is passed to the model as vectors; however, for image data, it is more helpful to have the data arranged as matrices because we want to capture the relationship of the pixel values in two-dimensional space.
Convolutional neural networks are able to capture these two-dimensional relationships through the use of a filter that convolves over the image data. The filter is a matrix with constant values and dimensions that are smaller than the image data. The constant values are multiplied by the underlying values and the sum of the resulting products is passed through to an activation function.
The activation function step, which can also be considered a separate layer, evaluates whether a given pattern is present in an...
