Blurb::
Non-hierarchical surrogates employ an unordered set of lower-fidelity
models to approximate a truth reference model at reduced cost.

Description::

Non-hierarchical surrogates employ an unordered set of lower-fidelity
model approximations to compute compute response quantities that are
correlated with truth reference values but can be obtained at reduced
cost.  Unlike the \c hierarchical surrogate, there is no implied
structure to this low-fidelity model ensemble other than there is a
single "truth" reference model that the rest of the models seek to
approximate.

The required \c unordered_model_fidelities specification points to a
set of specifications for model approximations and the \c
truth_model_pointer specification points to the high-fidelity "truth" 
model.  Each of these unordered model fidelities provides an
approximation to the truth model, and the reduced cost relative to
the truth model evaluations is specified as part of each model
specification (see \c solution_level_cost for simulation models).

Similar to the \c hierarchical surrogate, each of the approximation
models may contain resolution hyper-parameters which can be targets
for tuning the predictive value of each model approximation (see use
cases below).

Unlike the \c hierarchical surrogate, no \c correction specification is
used as no model discrepancies are employed.

<b> Use cases </b>

In the context of optimization, capabilities for low-fidelity model
tuning are under development.

In multifidelity uncertainty quantification (UQ), non-hierarchical
sampling approaches such as multifidelity Monte Carlo (MFMC) and
approximate control variate (ACV) sampling are now available.

In Bayesian inference, capabilities for low-fidelity model selection
are under development.

Description::


Topics::
Examples::
Theory::

Faq::
See_Also::	model-surrogate-hierarchical 
