A gaussian process models a function at discrete test points given training points and for each . Unlike other models, gaussian processes produce estimates for only the test points; however, their advantage lies in additionally providing the uncertainty of their estimates.
A gaussian process, at its core, models a multivariate 👑 Gaussian distribution over and together. We first start with a prior distribution, , which defines the model’s shape (periodic, linear or something else) without knowing the training data. Then, given training data , we can construct a posterior . TODO FINISH