Surrogate Methods

Synopsis: In engineering, surrogate models are often used to expedite the search for promising
designs by standing in for expensive design evaluations or simulations. Surrogates provide a fast
global model of a quantity of interest, such as aerodynamic drag or vehicle mass, which can then
be optimized efficiently. This course introduces the fundamentals of building, selecting,
validating, searching, and refining a surrogate model. Students will learn the theory behind the
surrogate methods as well as how to implement and apply them to simple and practical design
optimization problems. Course work includes homework assignments, tests, and a semester
project. The course should appeal to senior undergraduate students and graduate students.

Lecture topics include:

• Sampling plans
• Surrogate model construction, including Gaussian process regression and neural
networks
• Exploring and exploiting a surrogate
• Constraints
• Exploiting gradient information
• Multifidelity analysis
• Uncertainty analysis