Bayesian Optimization
Discover Factor Settings to Optimize Responses
The Bayesian Optimization platform is available only in JMP Pro.
The Bayesian Optimization platform uses an iterative learning technique to optimize one or more responses. This technique provides guidance on collecting additional data and when to stop collecting data. This platform can reduce the time and resources that are needed for many types of process and product development.
For each response, a Gaussian Process model is fit to an existing set of training data. This data set can come from either historical data or an initial set of experimental runs. The predictions from the Gaussian Process model, along with prediction uncertainty measures and response goals, are then used to create a set of acquisition functions. A space filling design is used to generate a set of new factor combinations, called a candidate set. Using manual or automated techniques, a set of new factor combinations is selected from the candidate set to be added to the original data. This set is referred to as a batch of runs. After new data are collected for this batch of runs, you relaunch the Bayesian Optimization platform to update the models and find a new batch of runs for consideration. This process can be repeated until the response goals are met and an optimal combination of factor settings is discovered.
Figure 18.1 Bayesian Optimization Report
