Gaussian Processes for Machine Learning
7. Gaussian Processes for Machine LearningΒΆ
Roughly speaking, Gaussian processes are a collection of random variables with Gaussian distributions. They are mainly characterized by their kernel or kernel function. In particular, they provide the foundation for probabilistic machine learning models belonging to the class of kernel-based methods. These methods use the kernel function to enable the use of a high-dimensional feature space. The purpose is to generate a more flexible machine learning model. This approach is particularly useful to generalize linear learning methods to non-linear settings and is also referred to as kernel trick. In this chapter, we focus on Gaussian processes, since they are most useful for our applications and fit into our probabilistic framework. First, let us discuss the intuition behind the kernels.