Abstract:Inspired by multi-fidelity methods in computer simulations, this article introduces procedures to design surrogates for the input/output relationship of a high-fidelity code. These surrogates should be learned from runs of both the high-fidelity and low-fidelity codes and be accompanied by error guarantees that are deterministic rather than stochastic. For this purpose, the article advocates a framework tied to a theory focusing on worst-case guarantees, namely Optimal Recovery. The multi-fidelity considerations triggered new theoretical results in three scenarios: the globally optimal estimation of linear functionals, the globally optimal approximation of arbitrary quantities of interest in Hilbert spaces, and their locally optimal approximation, still within Hilbert spaces. The latter scenario boils down to the determination of the Chebyshev center for the intersection of two hyperellipsoids. It is worth noting that the mathematical framework presented here, together with its possible extension, seems to be relevant in several other contexts briefly discussed.

What problem does this paper attempt to address?

The problem that this paper attempts to solve is: in the context of multi - fidelity models, design surrogate models that can learn from high - fidelity and low - fidelity code runs, and these surrogate models should have deterministic error guarantees rather than stochastic error guarantees. Specifically, the paper focuses on how to use the Optimal Recovery theory to construct surrogate models for multi - fidelity computer simulations and provide deterministic upper bounds on errors. This involves the following points: 1. **Combination of multi - fidelity data**: The paper explores how to combine the results of high - fidelity and low - fidelity models to better approximate the relationship between input and output and perform fine - tuning through the high - fidelity model. 2. **Deterministic error bounds**: Unlike existing Gaussian - process - based methods that can only provide stochastic error bounds, the method proposed in this paper aims to provide deterministic error bounds. This is very important for certain application scenarios because they require stricter error control. 3. **Solving optimization problems**: The paper proposes a framework for solving problems such as global optimal estimation of linear functionals, global optimal approximation of arbitrary quantities, and local optimal approximation. These problems ultimately boil down to determining the Chebyshev center of the intersection of two hyper - ellipsoids. ### Mathematical formulation Let the input / output relationship of the high - fidelity code be represented by the function \( f_0 \), and the relationship of the low - fidelity code be represented by \( f_1 \). The prior assumptions are formalized as: \[ f_0\in K_0,\quad f_0 - f_1\in K_1 \] where \( K_0 \) and \( K_1 \) are given sets, representing the constraints on \( f_0 \) and \( f_0 - f_1 \) respectively. Executing high - fidelity and low - fidelity codes provides data: \[ y_{0,i}=\lambda_{0,i}(f_0),\quad i = 1,\ldots,m_0 \] \[ y_{1,i}=\lambda_{1,i}(f_1),\quad i = 1,\ldots,m_1 \] where \( \lambda_{0,i} \) and \( \lambda_{1,i} \) are linear functionals, usually point evaluations. To find the optimal surrogate model \( h \), we need to minimize the worst - case full - approximation error, that is, solve the following minimax problem: \[ \inf_h\sup\{\| f_0 - h \| : f_0\in K_0\cap D_0, f_1 - f_0\in K_1, f_1\in D_1\} \] ### Application scenarios The method in the paper is applicable not only to multi - fidelity models but also to other fields, such as accelerated life testing, multi - physical sensing, panel data analysis, and transfer learning. These applications all require combining multiple related tests or data sets to improve prediction accuracy and provide deterministic error bounds. In summary, by introducing the Optimal Recovery theory, this paper solves the problem of surrogate model design in multi - fidelity models and provides deterministic error guarantees, thus providing new tools for the simulation and optimization of complex systems.

Worst-Case Learning under a Multi-fidelity Model

Multi-Fidelity Surrogate Based on Single Linear Regression

A survey on multi-fidelity surrogates for simulators with functional outputs: unified framework and benchmark

Multi-fidelity reduced-order surrogate modelling

Residual Multi-Fidelity Neural Network Computing

Hyperparameter-optimized Multi-fidelity Deep Neural Network model associated with Subset Simulation for structural reliability analysis

Multi-fidelity Machine Learning for Uncertainty Quantification and Optimization

Practical multi-fidelity machine learning: fusion of deterministic and Bayesian models

A Generalized Multi-Fidelity Simulation Method Using Sparse Polynomial Chaos Expansion

Multi-fidelity data fusion for the approximation of scalar functions with low intrinsic dimensionality through active subspaces

Graph Laplacian-based Bayesian Multi-fidelity Modeling

Multi-fidelity modelling via recursive co-kriging and Gaussian-Markov random fields

A multi-fidelity active learning method for global design optimization problems with noisy evaluations

Quantifying Learning Guarantees for Convex but Inconsistent Surrogates

A Single-Fidelity Surrogate Modeling Method Based on Nonlinearity Integrated Multi-Fidelity Surrogate

Multi-fidelity Gaussian process surrogate modeling for regression problems in physics

Multi-fidelity wavelet neural operator with application to uncertainty quantification

Recursive co-kriging model for Design of Computer experiments with multiple levels of fidelity with an application to hydrodynamic

Multifidelity linear regression for scientific machine learning from scarce data

Multi-Fidelity Adaptive Sampling for Surrogate-Based Optimization and Uncertainty Quantification

A comprehensive framework for multi-fidelity surrogate modeling with noisy data: a gray-box perspective