A GAN based solver idea for derivative-free optimization problems
Sprache des Titels:
Neural Information Processing Systems Foundation (NeurIPS 2019), 2019
We propose a GAN based approach for derivative-free function optimization.
The idea is to formulate the optimization process as an adversarial game where the generator has to propose new samples and the discriminator has to assess the quality of the samples with respect to a black-box function $f$, for which we do not have the analytical form but can only sample from. However, instead of attempting to approximate $f$ directly, the discriminator only has to solve a binary classification task in local regions populated by the generated samples. We demonstrate the efficacy of our approach by applying it to an artificially generated topology optimization problem. We show that, despite suppressing access to derivatives of $f$, our method leads to similar results like more traditional topology optimization methods for which have access to derivatives. We hypothesize that our approach is a potential neural counterpart to derivative-free optimization methods.