Developer Tutorial: Bayesian Optimization

This session is for R&D product and process engineers responsible for innovating quickly to be the first to market with a new product

Bayesian Optimization is an iterative learning technique that uses a model to generate new candidate runs for assessment. Bayesian Optimization learns from the responses with each iteration and gives us clearer guidance than traditional approaches about when we can stop experimenting. The iterative nature has the potential to dramatically reduce the time and resources required for process and product development, so may be a natural fit for many R&D cultures.

Starting with existing data, either historical or from an initial set of experimental runs, a Gaussian Process model (GaSP) is constructed. Then, using the model predictions, along with the prediction uncertainty and the response goals, new candidate factor combinations are generated to test. After test, the model is updated with the new data. This sequence of steps repeats until the response goals are met and an optimal combination of factor settings is discovered.

New enhancement to JMP Pro incorporate an interactive new platform for Bayesian Optimization.

JMP Chief Data Scientist and a key Developer of the new Bayesian Optimization, Chris Gotwalt, will demonstrate and explain the capability and the underlying statistical approaches is deploys. The session includes time for Q&A.

This JMP Developer Tutorial covers: interactively generating the initial Gaussian Process model; specifying new candidate factor combinations to test or run; exploring design space; summarize Gaussian Process model fit; comparing and interactively doing what-if analyses on factors and responses.