CVPP - Bayesian Optimization
Bayesian Optimization (BO) is a sequential technique for the global optimization of black-box functions, that are either too complex or expensive to be optimized directly. It treats the objective function as unknown and places a prior distribution over it, which captures our beliefs about its behavior. The objective function is evaluated and the observations are used to produce the posterior distribution. This posterior distribution is then used to construct an acquisition function that determines where the next query point should be, repeating the cycle.
The CVPP library contains various implementations of Bayesian Optimization algorithms, and is directly connected to the Gaussian Process module, since it uses a GP to model stored observations and determine the next query point. Additionally, it provides most of the commonly used acquisition functions, alongside a template that facilitates the creation of customized ones.
Below are some demos of BO models already available in the CVPP library, with brief explanations, references and videos.
The CVPP library contains various implementations of Bayesian Optimization algorithms, and is directly connected to the Gaussian Process module, since it uses a GP to model stored observations and determine the next query point. Additionally, it provides most of the commonly used acquisition functions, alongside a template that facilitates the creation of customized ones.
Below are some demos of BO models already available in the CVPP library, with brief explanations, references and videos.
BayesOpt is the term used by CVPP when referring to the standard Discrete Bayesian Optimization derivation, in which only the goal point is taken into consideration when determining the next queries at each iteration.
|
|