Bayesian Optimization: expected improvement, example (+LCB,PI,EV)

Antonio Sala, UPV

Difficulty: **** ,       Relevance: ,      Duration: 19:58

*Link to my [ whole collection] of videos in English. Link to larger [ Colección completa] in Spanish.

Abstract/Summary:

This video, a continuation of [boPIEN], will discuss the meaning of the Expected Improvement (EI) acquisition function in Bayiesan optimization. Basically, Expected value (Gaussian mean) should be used if getting a bad sample penalises some “utility” in my particular application, and expected improvement should be used if bad samples do not incur any cost as we have our ‘best’ sample in our historical record to provide as our final solution to the optimization problem.

A final discussion on the comparison with probability of improvement and confidence bound acquisition functions is also provided.

All discussions are supported via an example that is coincident with that in the above-mentioned video this is a continuation thereof.