Improving Evolutionary Strategies with Generative Neural Networks
This work addresses optimization efficiency for researchers and practitioners using ES algorithms, but it appears incremental as it builds on existing ES methods with a novel distribution modeling approach.
The paper tackled the problem of improving Evolutionary Strategies (ES) by using Generative Neural Networks (GNNs) to model flexible search distributions, resulting in a new training algorithm that outperforms state-of-the-art methods on diverse objective functions.
Evolutionary Strategies (ES) are a popular family of black-box zeroth-order optimization algorithms which rely on search distributions to efficiently optimize a large variety of objective functions. This paper investigates the potential benefits of using highly flexible search distributions in classical ES algorithms, in contrast to standard ones (typically Gaussians). We model such distributions with Generative Neural Networks (GNNs) and introduce a new training algorithm that leverages their expressiveness to accelerate the ES procedure. We show that this tailored algorithm can readily incorporate existing ES algorithms, and outperforms the state-of-the-art on diverse objective functions.