Introduction to AI Pt3 - Stacking neurons into layers; calculus intuition and cost functions

In this video we introduce a computer vision problem, that of identifying items of clothing from small, monochrome images. We look at how we would design a neural network to solve this problem. This involves stacking neurons into layers, and feeding information forward through stacks of neurons until we reach an output layer. This output layer has as many neurons as items of clothing that we are looking for. In this way each output neuron in our output layer 'looks' for a particular item of clothing. This is to say its activation is highest when the image contains the item of clothing that it has been trained to look for. We also look at the application of cost functions - our 'coaches' for our neural networks. We see how these work back-to-front, from output layer backwards to input layer. As the const function operates from back to front it makes small nudges to the network's parameters to enable the network to become more accurate while training