{"about":{"site":"https://codewithpapers.app","non_affiliation":"Code with Papers and Syntology are not affiliated with, endorsed by, or sponsored by Papers with Code, Meta, or the pwc-archive mirror.","licence":"CC BY-SA 4.0","licence_url":"https://creativecommons.org/licenses/by-sa/4.0/legalcode","attribution":"https://codewithpapers.app/attribution","modified":"archive material modified by Syntology; see the attribution page"},"url":"/paper/the-numerics-of-gans","title":"The Numerics of GANs","arxiv_id":"1705.10461","date":"2017-05-30","proceeding":"NeurIPS 2017 12","authors":["Lars Mescheder","Sebastian Nowozin","Andreas Geiger"],"abstract":"In this paper, we analyze the numerics of common algorithms for training\nGenerative Adversarial Networks (GANs). Using the formalism of smooth\ntwo-player games we analyze the associated gradient vector field of GAN\ntraining objectives. Our findings suggest that the convergence of current\nalgorithms suffers due to two factors: i) presence of eigenvalues of the\nJacobian of the gradient vector field with zero real-part, and ii) eigenvalues\nwith big imaginary part. Using these findings, we design a new algorithm that\novercomes some of these limitations and has better convergence properties.\nExperimentally, we demonstrate its superiority on training common GAN\narchitectures and show convergence on GAN architectures that are known to be\nnotoriously hard to train.","url_abs":"http://arxiv.org/abs/1705.10461v3","url_pdf":"http://arxiv.org/pdf/1705.10461v3.pdf","source":{"archive":"pwc-archive (Hugging Face), CC BY-SA 4.0","snapshot":"2025-07-28","licence_url":"https://creativecommons.org/licenses/by-sa/4.0/legalcode","row_kind":"abstracts"},"code_links":[{"paper_slug":"the-numerics-of-gans","repo_url":"https://github.com/LMescheder/TheNumericsOfGANs","is_official":1,"mentioned_in_paper":1,"mentioned_in_github":0,"framework":"tf","reach":{"status":"unanswered"}},{"paper_slug":"the-numerics-of-gans","repo_url":"https://github.com/Phutoast/Stable-Multi-Agent-Reinforcement-Learning","is_official":0,"mentioned_in_paper":0,"mentioned_in_github":1,"framework":"pytorch","reach":{"status":"unanswered"}},{"paper_slug":"the-numerics-of-gans","repo_url":"https://github.com/nhynes/abc","is_official":0,"mentioned_in_paper":0,"mentioned_in_github":1,"framework":"pytorch","reach":{"status":"unanswered"}},{"paper_slug":"the-numerics-of-gans","repo_url":"https://github.com/limcherhang/CNCO","is_official":0,"mentioned_in_paper":0,"mentioned_in_github":0,"framework":"pytorch","reach":{"status":"unanswered"}}],"tasks":[],"methods":[{"method_slug":"convolution","method_name":"Convolution"}],"datasets_introduced":[],"methods_introduced":[],"results":[],"syntology":{"atlas_url":"https://app.syntology.ai/?focus=1705.10461","mcp":null,"developers":"https://syntology.ai/developers"},"arxiv_metadata":null,"syntology_extracted_results":null}