Building neural networks for MNIST digit recognition; activation, cost, and backpropagation.

Using a single neuron to perform linear regression and train via gradient descent.

Overview of artificial neuron models and activation functions (sigmoid, tanh, ReLU).