Flashcards
Bayesian Optimisation
Cards in this deck (3)
- 1What is BO?Specifying a prior to ask the black box and query the function at different locations, get a posterior distribution over what f might be based on previous data
- 2How to do BO?Use GPs so we have uncertainty and an acquisition function to tell us the benefit to query f at a specific point to get the minima.
Find acquisition function maxima and update the posterior
Formulate prior of objective function f(x)
Query objective function
Derive posterior distribution using known values of the objective function
Evaluate acquisition function and find best place
Go to 2 and repeat - 3How to get a good acquisition function?Balances exploration of space x to gain new knowledge and exploitation of previous knowledge
Information
- Description
- Machine Learning Flashcards on Bayesian Optimisation, created by Philip Masters on 03/01/2020.
- Created by
- Philip Masters
- Views
- 1
- Rating