SMTorg/smt

What it solves

SMT provides a collection of surrogate modeling methods, sampling techniques, and benchmarking functions. It is designed to replace expensive computational simulations with fast, approximate models (surrogates) to accelerate engineering design and analysis.

How it works

The toolbox implements various surrogate models, including Gaussian Processes (Kriging), neural networks, and spline interpolation. A key differentiator is its heavy emphasis on derivatives; it supports training derivatives for gradient-enhanced modeling, prediction derivatives, and derivatives with respect to training data.

Who it’s for

Engineers and researchers in fields like aerodynamics and structural optimization who need to build fast approximations of complex, high-dimensional simulation data.

Highlights

• Derivative-focused: Specifically designed for gradient-enhanced modeling and prediction derivatives.
• Unique Models: Includes specialized methods like kriging by partial-least squares reduction and energy-minimizing spline interpolation.
• Hierarchical & Mixed Variables: Focuses on Gaussian Processes for hierarchical and mixed-variable spaces.
• Extensible: Simple library structure that facilitates the implementation of new surrogate methods.