The Deep Learning Architect's Handbook

3

Understanding Convolutional Neural Networks

An MLP is structured to accept one-dimensional data and cannot directly work with two-dimensional data or higher-dimensional data without preprocessing. One-dimensional data is also called tabular data, which commonly includes categorical data, numerical data, and maybe text data. Two-dimensional data, or data with higher dimensions, is some form of image data. Image data can be in two-dimensional format when it is a grayscale formatted image, in three-dimensional format when it has RGB layers that closely represent what humans see, or in more than three dimensions with hyperspectral images. Usually, to make MLP work for images, you would have to flatten the data and effectively represent the same ...