definition · term 22 of 25
surrogate model.
A fast machine learning model trained to imitate a slow simulation or a system known only from data.
worked exampleseconds instead of hoursA process simulation takes an hour per run. A surrogate trained on 500 runs answers in milliseconds, so the optimizer can try thousands of settings and check the best few in the simulator.
in plain language
When each evaluation of a system takes hours of simulation or a real production run, optimization cannot query it thousands of times. A surrogate model, such as a gradient-boosted tree ensemble or a Gaussian process, learns the input-output relation. It answers in milliseconds, with an estimate of its own uncertainty.
surrogate model, on your own decision.
Bring one recurring decision. Your planners see the terms of this glossary applied to their own rules and data.
- Glossary
- all 25 terms, a to z
- Problem types
- the decisions kint solves
- Comparisons
- kint next to other approaches
- Contact
- hello@kint.pro