Data Dieting in GAN Training
This addresses data efficiency in GAN training for researchers and practitioners, but it is incremental as it builds on existing methods like Redux-Lipizzaner.
The paper tackles the problem of training Generative Adversarial Networks (GANs) with reduced data to lower resource requirements, finding that subsets can maintain sample diversity and that generator ensembles add value, with experiments on MNIST and CelebA datasets.
We investigate training Generative Adversarial Networks, GANs, with less data. Subsets of the training dataset can express empirical sample diversity while reducing training resource requirements, e.g. time and memory. We ask how much data reduction impacts generator performance and gauge the additive value of generator ensembles. In addition to considering stand-alone GAN training and ensembles of generator models, we also consider reduced data training on an evolutionary GAN training framework named Redux-Lipizzaner. Redux-Lipizzaner makes GAN training more robust and accurate by exploiting overlapping neighborhood-based training on a spatial 2D grid. We conduct empirical experiments on Redux-Lipizzaner using the MNIST and CelebA data sets.