I am still new to CNNs, but I would like to check my understanding between when to use convolutional layers versus fully connected layers.
From what I have read, we can use convolutional layers with filters, rather than fully connected layers, with images, text, and audio. However, with regular data, for example, the iris dataset, a convolutional layer would not perform well because of the structure. As in the columns can be swapped, yet the record or sample itself does not change. For example we can swap the order of the Petal Length column with Petal Width and the record does not change. Where as in an image or audio, changing the column items would result in a different image or audio file.
These convolutional layers are "better" for images and audio because not all the features need to connect to the next layer. For example, we do not need the background of a car image to know it is a car, thus we do not need all the connections and we save computational costs.
Is this the right way to think about when to use convolutional layers versus fully connected layers?
