Conventional gaussian process regression exclusively assumes the existence of
noise in the output data of model observations. In many scientific and
engineering applications, however, the input locations of observational data
may also be compromised with uncertainties owing to modeling