Sean Schweiger

Global optimization serves as a cornerstone in various scientific disciplines, playing a pivotal role in solving complex problems and refining models crucial for understanding the world around us. However, global optimization problems can vary greatly in nature, making them hard to solve without expert knowledge. Our research revolves around the idea of using deep neural networks to reparameterize optimization problems, which we call Deeplifting.This would effectively “lift” the problem to a much higher dimension, which we hypothesize would result in a smoother optimization landscape and lead to more optimal solutions. To study our approach, we employed our method on an array of problem instances that we compiled from several widely used lists for benchmarking global optimization solvers. We compared the results of our method to the results of several traditional global optimization solvers. In our findings, we found that the global solution was found at a higher rate for many of the test problems using our Deeplifting method.