Sequential Design of Experiments Using Bayesian OptimizationI have a customer who asked to learn more about this topic. He read a paper recently that talked about this approach to DOE: "Design of Biopharmaceutical Formulations Accelerated by Machine Learning" (attached here).
Is this a topic anyone is familiar with? If so - I would like to discuss with you.
It appears that @yuichi_katsumur has an add-in (Bayesian optimization add-in) that he built to support this methodology and I have reached out to him but curious if others have come across this.
Re: Sequential Design of Experiments Using Bayesian OptimizationJust to chime in, we are aware of this in R&D.
I'm working on a project with @Wenjun_Bao_JMP that ties into the XGBoost platform for sequential experimentation, particularly with a view to working with the new Autotune feature @russ_wolfinger has added for JMP 17.
This would be more of a "batch sequential" approach that is trying both exploitation and exploration, instead of more of the one-experiment-at-a-time exploitation.
Cheers,
Ryan
Re: Sequential Design of Experiments Using Bayesian OptimizationHi @wendytseng,
I had some experience with Bayesian optimization in my previous role and know someone doing research in the field of BO. His name is Prof. Marius Lindauer. We actually tried to get funding for a project to compare known DoE approaches with the BO approach in process development. Seems to be of rising interest in research as far as I know.
Best,
Jonas
Re: Sequential Design of Experiments Using Bayesian OptimizationHi @wendytseng ,
You have probably already seen this older post about this same topic in the SE Discussions. (Including it for future reference).
There does seem to be increasing interest in this type of approach. It could be a good topic for a technical intern to explore.
Cheers,
Phil
Re: Sequential Design of Experiments Using Bayesian OptimizationHi @wendytseng ,
Thank you for reaching out to me. I created the add-in to support the methodology you mentioned.
Some customers in Japan are interested in this approach. I think Bayesian optimization is suitable for simulation experiments because Bayesian optimization uses a Gaussian process model for making a surrogate model. If your customer try to use Bayesian optimization for actual experiments, I recommend starting from a small experiment to see whether Bayesian optimization works in their experiments.
