Generative Adversarial Network (GAN) is a family of machine learning algorithms designed to train neural networks able to imitate real data distributions. Unfortunately, GAN suffers from problems such as gradient vanishing and mode collapse. In Multi-Objective Evolutionary Generative Adversarial Network (MO-EGAN) these problems were addressed using an evolutionary technique combined with Multi-Objective selection, obtaining better results on synthetic datasets at the expense of larger computation times. In this works, we present the Smart Multi-Objective Evolutionary Generative Adversarial Network (SMO-EGAN) algorithm, which reduces the computational cost of MO-EGAN and achieves better results on real data distributions.
Smart Multi-Objective Evolutionary GaN
Baioletti M.;Di Bari G.;Poggioni V.;
2021
Abstract
Generative Adversarial Network (GAN) is a family of machine learning algorithms designed to train neural networks able to imitate real data distributions. Unfortunately, GAN suffers from problems such as gradient vanishing and mode collapse. In Multi-Objective Evolutionary Generative Adversarial Network (MO-EGAN) these problems were addressed using an evolutionary technique combined with Multi-Objective selection, obtaining better results on synthetic datasets at the expense of larger computation times. In this works, we present the Smart Multi-Objective Evolutionary Generative Adversarial Network (SMO-EGAN) algorithm, which reduces the computational cost of MO-EGAN and achieves better results on real data distributions.I documenti in IRIS sono protetti da copyright e tutti i diritti sono riservati, salvo diversa indicazione.