Bayesian Optimization motivation (2/4): methodology outline

Antonio Sala, UPV

Difficulty: *** ,       Relevance: ,      Duration: 11:20

Summary:

This video continues with the introduction and motivation to Bayesian Optimization problems which was started in video [BOmot1EN].

In here, we review said video in the first four minutes, and then we discuss the generic goal of Bayesian Optimization: it’s actually a type of ‘experiment design’ trying to choose samples that have a high likelihood of giving me a value close to the optimal (exploitation) or, well, maybe we wish to sample currently likely sub-optimal ones in exchange for them giving a lot of information on where the actual optimum will be for future samples (exploration).

So, BO algorithms end up comprising the following steps:

• Set up a prior statistical model

• Acquire samples (this may be a costly procedure in some relevant application cases)

• Compute a posterior (conditional to the measurements)

• Carry out a statistical analysis on the posterior, to decide next sample, and go to step 2, unless some termination criteria (sample budged, quality of the last sample, lack of progress) is met.

For brevity, a quick outline of the details of the third and fourth steps will be deferred to video [BOmot3EN], and application cases and concluding remarks in video [BOmot4EN] will conclude this brief motivation and presentation of the Bayesian optimization problem. A quick numerical example of the methodology appears in video [boloop1EN], but you need to watch [BOmot3EN] first to better understand it.

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