The strategy that the dlib implementation uses of combining a global optimization method (LIPO) with one that works better locally (TR) reminds me of another re

The strategy that the dlib implementation uses of combining a global optimization method (LIPO) with one that works better locally (TR) reminds me of another recent paper I ran across, which takes a black-box global optimization algorithm (StoSOO) and improves local convergence behavior using CMA-ES [1]. A main difference seems to be that StoSOO focuses on robustness to noisy functions more than LIPO does. [1] https://link.springer.com/article/10.1007/s10898-016-0482-9 https://link.springer.com/article/10.1007/s10898-016-0482-9 / http://ssamot.me/papers/global-optimisation.pdf http://ssamot.me/papers/global-optimisation.pdf