Abstract
The machine learning generative algorithms such as Generative AdversarialNetworks (GAN) and Variational Auto-Encoders (VAE) show impressive results whenconstructing objects similar to those in a training ensemble. However, thegeneration of new objects builds mainly on the understanding of the hiddenstructure of the training dataset followed by a sampling from amulti-dimensional normal variable. In particular each sample is independentfrom the others and can repeatedly propose same type of objects. To cure thisdrawback we introduce a kernel-based measure quantization method that canproduce new objects from a given target measure by approximating it as a wholeand even staying away from elements already drawn from that distribution. Thisensures a better diversity of the produced objects. The method is tested onclassic machine learning benchmarks.