NeurIPS 2020poster65 citations

Top-k Training of GANs: Improving GAN Performance by Throwing Away Bad Samples

Samarth Sinha, Zhengli Zhao, Anirudh Goyal ALIAS PARTH GOYAL, Colin A Raffel, Augustus Odena

Abstract

We introduce a simple (one line of code) modification to the Generative Adversarial Network (GAN) training algorithm that materially improves results with no increase in computational cost. When updating the generator parameters, we simply zero out the gradient contributions from the elements of the batch that the critic scores as

BibTeX
@inproceedings{NEURIPS2020_a851bd0d,
 author = {Sinha, Samarth and Zhao, Zhengli and ALIAS PARTH GOYAL, Anirudh Goyal and Raffel, Colin A and Odena, Augustus},
 booktitle = {Advances in Neural Information Processing Systems},
 editor = {H. Larochelle and M. Ranzato and R. Hadsell and M.F. Balcan and H. Lin},
 pages = {14638--14649},
 publisher = {Curran Associates, Inc.},
 title = {Top-k Training of GANs: Improving GAN Performance by Throwing Away Bad Samples},
 url = {https://proceedings.neurips.cc/paper_files/paper/2020/file/a851bd0d418b13310dd1e5e3ac7318ab-Paper.pdf},
 volume = {33},
 year = {2020}
}