Bespoke Learning in Static Systems: Application to Learning Sub-surface Material Density Function

In this chapter we discuss bespoke learning of a system property in a static system, i.e. discuss bespoke learning of the relevant system property parameter, at design values of an input. Such learning will then capacitate the supervised learning of the system property as a function of the input variable. This bespoke learning typically requires anticipating trends in values of the likelihood, as it changes with the difference between observations and (a model of) the system. Modelling the likelihood in terms of such a difference—or distance—is possible, subsequent to mapping from the space of system parameters onto the space of the observable. Here, we discuss an application of this scheme, to the non-destructive learning of the material density, as a function of a sub-surface location—using existing work that provides the bespoke learnt values of this density at designed sub-surface locations. The training data that is thus generated, bears strong inhomogeneities in correlation, rendering supervised learning of the material density function, difficult. This is undertaken, to enable prediction of the density at test locations.