Figure 2
From: Efficient Closed-loop Maximization of Carbon Nanotube Growth Rate using Bayesian Optimization
An illustration of Bayesian optimization applied to a toy problem using the Matern52 kernel and the upper confidence bound acquisition function. In each plot, a GP is represented by its mean function (thick green line) and standard deviation (blue area), along with ten random functions sampled from this GP (thin green lines). BO begins with an initial rough approximation (i.e., prior) about the underlying unknown function and then refines the approximation trial by trial as new observations are made. Observations (black dots) are collected from the ground truth function (red line). At any given trial, BO selects the point with highest acquisition function (dashed line) to evaluate, a new observation (red dot) is made at the next trial, and this sequential adaptive process then repeats itself.
