32. Bayesian Optimization

[@franciscochagas-hp8yy]

professor, i'm study industrial chemistry in a university on brazil, and i go to learn more about this, i think is so much interesting in the chemical industry.
i learn about this in other articles provide in cambridge, is so usefull

[@TaylorSparks]

@@franciscochagas-hp8yy it is such a great topic. Best of luck in learning it and applying it to your research

[@jinlongsu7308]

An excellent video. Thank you so much professor! 😀May I ask that if I want to optimise two or more properties together (e.g., strength and ductility), would it be better if I use a Multi-task Gaussian Process (MTGP) model, rather than several individual GP models? Because the trade-off between these properties can be learned by the MTGP model? By the way, may I check that can the iteration of the bayesian optimisation process also be termed as Active Learning, if I am not wrong? Many thanks for your attention. 🙏

[@TaylorSparks]

@@jinlongsu7308 I would simply do a multi objective optimization. Multitask is typically used when one property has more statistical strength than the other and the two properties are related their by allowing you to lean on the statistical strength of the one with the larger data set

[@jinlongsu7308]

@@TaylorSparks Noted. Many thanks for your prompt reply! 😀

[@vrhstpso]

Thank you professor 🙏. If we knew the objective function ( for example mean squared error equation) would it be logical to make use of bayesian optimization to find the minumum of that objective function which is computed for aimed optimized parameters in a given domain?

[@TaylorSparks]

@@vrhstpso sure, but it is more often the case that we don't know the objective function and we are instead trying to learn it using a surrogate model such as a gaussian process.

[@vrhstpso]

@@TaylorSparks thank you.