Reliable probabilistic supervised learning, given small to large-sized training sets, for fast closed-form prediction at test inputs.
In various disciplines, we often seek a data-driven and reliable learning of a generic high-dimensional function that represents the relationship between a vector input and a generally high-dimensional output.
The ultimate aim is the closed-form and fast prediction at a test input, though some applications demand the learning of the input at which a new value of the output has been realised, where said (inverse) input learning helps circumvent the problem of smallness of training set.
Such probabilistic mechanistic learning has been undertaken, of the relation between disease severity score and pre-onset characteristics of the patient, to perform automated prediction of disease severity score of multiple prospective patients. An AI capacity has been developed to perform the prediction.
Again, such supervised learning is undertaken – using two nested layers of Gaussian Processes – of the relationship between the vector of Milky Way feature parameters, and the matrix of velocity components of stellar neighbours of the Sun, to predict the galactic features at which an observed velocity matrix of such solar neighbours is realised.