Machine Learning But Funner 02 - The (Simplified) Theory of Convolutional Neural Networks
Today I aim to cover the basic theory of convolutional neural networks in the most high level fashion I can, ignoring (or as we computer scientists like to lie to ourselves: abstracting away from) most of the maths, targeting this tutorial at complete beginners. However I will mention as an obligatory disclaimer now that in order to gain a deeper understanding of the topic i.e. to build a CNN from scratch, I will do as every blogger and direct you to an over-detailed book: Artificial Intelligence for Humans Volume 3: Deep Learning and Neural Networks which covers the topic in Chapter 10. Network Structure The CNN model is built on top of the usual feed-forward neural network model and so the structure is similar to what you may be comfortable with already, except the 'hidden layer' will be replaced by the layers below: Convolutional Layers Pooling Layers (here we will use max-pooling but there are others) Dense Layers (this is often the same as a regular output laye