Here we are going to look at linear models, or to be more precise general linear models.

They might not be the sexiest of machine learning models, but they provide the foundations for many more sophisticated algorithms which include deep neural networks and Gaussian Processes, which you will learn about throughout more of the digiLab academy courses. They also give us the stage to talk about certain key principals in machine learning, like overfitting and regularisation in a simplier model setting.

We'll cover the following topics:

• Step back to curve fitting as a general process, with straight lines and polynomials in general.

• Talk about General Linear Models, how do we fit general linear model made up of general features and see how these can be very rich and expressive models.

• Understand the core concetps of underfitting and overfitting, and then how do we overcome these challenges in particular by using regularisation.

• Talk about approaches for discovering relationships in data by looking at techniques for linear and nonlinear relationships.

• Work through our first unsupervised learning approach for dimension reduction, Principle Component Analysis. Which fits with the view of linear models, since it is a linear transform of our input data.

• Finally we discover Logistic Regression, in particular how we can extend the linear models for regression tasks for classification problems.