Glossary term
Glossary term
Foundations
A probabilistic regression model technique for optimizing computationally expensive objective functions by instead optimizing a surrogate that quantifies the uncertainty using a Bayesian learning technique. Since Bayesian optimization is itself very expensive, it is usually used to optimize expensive-to-evaluate tasks that have a small number of parameters, such as selecting hyperparameters.
Created for this library
A computer vision team uses Bayesian optimization to tune learning rate, batch size, and dropout faster than grid search on expensive training runs.
A pharma research team uses Bayesian optimization to choose the next experiment when each in-vitro test takes a week and costs thousands of dollars.
An advertising platform team uses Bayesian optimization to tune the auction parameters of its bidding model when each evaluation requires a live A/B test.
Definition source: Google for Developers Machine Learning Glossary | Creative Commons Attribution 4.0 License