Key Concepts

What are kernels?

[TODO: mathematical functions with nice properties (PSD), allowing the computation of full mathematical objects like covariances from only limited parameters called “hyper-parameters”, which are often nicely interpretable.]

What is the kernel trick?

[TODO: explain]

When is it useful?

[TODO: this makes a lot of linear models able to work non-linearly, and is common in probabilistic models like GPs]

Families of kernels

[TODO: define (an)isotropic, stationnary, …]

[TODO: redirect to all_kernels.md and the kernel cookbook https://www.cs.toronto.edu/~duvenaud/cookbook/]

Kernel composition

[TODO: composing kernels enables fine signal properties identification, like de Mauna Loa CO2 emissions example]