In this course we’ll use the Machine Intelligence library TensorFlow to build an application that can detect handwritten numbers. The approach we’ll take is to train a neural network with thousands of photos of handwritten numbers, so that it can learn the right patterns and be able to recognize numbers on its own. As we train our neural network, we’ll reach a recognition accuracy of over 90%.
We’ll cover neural networks from scratch, starting with modeling a single neuron using the Perceptron model, which is similar to real neuron cells in the brain. We’ll move on from that to cover systems of multiple neurons and cover the Gradient Descent and Backpropagation techniques to train our networks.
To implement our handwriting recognition neural network we’ll be using the popular TensorFlow framework created by Google. We’ll visualize the training using TensorBoard.