We will show how to globally optimize multivariable functions on the GPU by exploiting Julia's unique capabilities in that area, using an algorithm of branch-and-bound type. We obtain up to 100x speedups on hard instances over a single CPU processor.
We will show how to obtain fast global optimization of difficult
unconstrained and constrained nonlinear programming problems in low and medium
dimension, using branch-and-bound type, which is naturally highly parallelizable. To do so, we exploit some of Julia's unique features to transfer algorithms from the CPU over to the GPU. We will discuss some of those features in detail and show how to combine them into different algorithms.