Tutorial: Bayesian Optimization

1D and 2D black-box Bayesian optimization demonstration with visualizations.

This code shows a visualization of each iteration in Bayesian Optimization. MATLAB's fitrgp is used to fit the Gaussian process surrogate model, then the next sample is chosen using the Expected Improvement acquisition function. An exploitation-exploration parameter can be changed in the code. The code contains both 1D and 2D "black-box" functions for optimization.

References:

[1] Rasmussen and Williams (2006). "Gaussian Processes for Machine Learning," MIT Press.

[2] Frazier (2018). https://arxiv.org/abs/1807.02811

[3] Snoek (2012). https://arxiv.org/pdf/1206.2944.pdf

[4] http://krasserm.github.io/2018/03/21/bayesian-optimization/