(4/5) Multi-output Gaussian Processes: joint prediction

Antonio Sala, UPV

Difficulty: **** ,       Relevance: ,      Duration: 17:42

Summary:

This video discusses prediction (Gaussian Process regression) in multi-output stochastic Gaussian processes, see video [mimoGP1EN] for motivation and introduction.

Actually, with the representation as a process of a single output with an input domain expanded to $f(x,i)$ with $i$ being a categorical ‘index’ or ‘label’ variable, prediction in an abstract sense is identical to the usual GP prediction, so there is nothing new from a theoretical point of view. Therefore, this video simply has a didactic objective: to understand how by measuring one variable another can be better estimated in close abscissa, complementing the case study of the videos [mimogp2EN] and [mimogp3pcaEN].

If the ‘latent’ factors $u$ in the statistical model had a physical meaning so its estimation seemed of interest in any particular application, that estimation is carried out in video [mimogp5predUEN], which closes the case study.

*Link to my [ whole collection] of videos in English. Link to larger [ Colección completa] in Spanish.