Bayesian optimization

E899020

black-box optimization technique global optimization method sample-efficient optimization method sequential model-based optimization

Bayesian optimization is a sample-efficient global optimization strategy that uses probabilistic surrogate models, typically Gaussian processes, to optimize expensive black-box functions with as few evaluations as possible.

Try in SPARQL Jump to: Statements Referenced by

Statements (48)

Predicate	Object
instanceOf	black-box optimization technique ⓘ global optimization method ⓘ sample-efficient optimization method ⓘ sequential model-based optimization ⓘ
aimsTo	minimize the number of function evaluations ⓘ optimize expensive black-box functions ⓘ
appliesTo	continuous optimization problems ⓘ experimental design ⓘ hyperparameter optimization in machine learning ⓘ mixed discrete-continuous optimization problems ⓘ simulation-based optimization ⓘ
assumes	function evaluations are expensive ⓘ function evaluations may be noisy ⓘ
basedOn	Bayesian inference NERFINISHED ⓘ sequential decision making ⓘ
challenge	parallel and batch evaluations design ⓘ scaling to high-dimensional problems ⓘ
commonAcquisitionFunction	entropy search ⓘ expected improvement ⓘ knowledge gradient ⓘ probability of improvement ⓘ upper confidence bound ⓘ
commonSurrogateModel	Bayesian neural networks NERFINISHED ⓘ Gaussian process regression ⓘ random forests ⓘ
contrastsWith	gradient-based optimization methods ⓘ grid search ⓘ random search ⓘ
handles	black-box objectives without analytic gradients ⓘ
models	posterior distribution over objective functions ⓘ
oftenAssumes	low-dimensional search spaces ⓘ
originField	machine learning ⓘ operations research ⓘ statistics ⓘ
property	global search capability ⓘ handles noisy observations ⓘ non-convex optimization capability ⓘ sample efficiency ⓘ
relatedTo	active learning ⓘ multi-armed bandits ⓘ optimal experimental design ⓘ
requires	likelihood model for observations ⓘ prior over functions ⓘ
selects	next evaluation point by maximizing an acquisition function ⓘ
updates	surrogate model with new observations ⓘ
uses	Gaussian processes NERFINISHED ⓘ acquisition functions ⓘ probabilistic surrogate models ⓘ

Referenced by (1)

Full triples — surface form annotated when it differs from this entity's canonical label.

AutoML: A Survey of the State-of-the-Art → topic → Bayesian optimization ⓘ