{"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/pacgan-the-power-of-two-samples-in-generative","title":"PacGAN: The power of two samples in generative adversarial networks","arxiv_id":"1712.04086","date":"2017-12-12","proceeding":"NeurIPS 2018 12","authors":["Zinan Lin","Ashish Khetan","Giulia Fanti","Sewoong Oh"],"abstract":"Generative adversarial networks (GANs) are innovative techniques for learning\ngenerative models of complex data distributions from samples. Despite\nremarkable recent improvements in generating realistic images, one of their\nmajor shortcomings is the fact that in practice, they tend to produce samples\nwith little diversity, even when trained on diverse datasets. This phenomenon,\nknown as mode collapse, has been the main focus of several recent advances in\nGANs. Yet there is little understanding of why mode collapse happens and why\nexisting approaches are able to mitigate mode collapse. We propose a principled\napproach to handling mode collapse, which we call packing. The main idea is to\nmodify the discriminator to make decisions based on multiple samples from the\nsame class, either real or artificially generated. We borrow analysis tools\nfrom binary hypothesis testing---in particular the seminal result of Blackwell\n[Bla53]---to prove a fundamental connection between packing and mode collapse.\nWe show that packing naturally penalizes generators with mode collapse, thereby\nfavoring generator distributions with less mode collapse during the training\nprocess. Numerical experiments on benchmark datasets suggests that packing\nprovides significant improvements in practice as well.","url_abs":"http://arxiv.org/abs/1712.04086v3","url_pdf":"http://arxiv.org/pdf/1712.04086v3.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":"pacgan-the-power-of-two-samples-in-generative","repo_url":"https://github.com/fjxmlzn/PacGAN","is_official":1,"mentioned_in_paper":1,"mentioned_in_github":1,"framework":"tf","reach":{"status":"unanswered"}},{"paper_slug":"pacgan-the-power-of-two-samples-in-generative","repo_url":"https://github.com/alex98chen/testGAN","is_official":0,"mentioned_in_paper":0,"mentioned_in_github":1,"framework":"tf","reach":{"status":"unanswered"}},{"paper_slug":"pacgan-the-power-of-two-samples-in-generative","repo_url":"https://github.com/bhargavajs07/Packed-Wasserstein-GAN-with-GradientPenalty-Example","is_official":0,"mentioned_in_paper":0,"mentioned_in_github":1,"framework":"pytorch","reach":{"status":"unanswered"}},{"paper_slug":"pacgan-the-power-of-two-samples-in-generative","repo_url":"https://github.com/bhargavajs07/Packed_WGAN_GP_Example","is_official":0,"mentioned_in_paper":0,"mentioned_in_github":1,"framework":"pytorch","reach":{"status":"unanswered"}},{"paper_slug":"pacgan-the-power-of-two-samples-in-generative","repo_url":"https://github.com/fjxmlzn/DoppelGANger","is_official":0,"mentioned_in_paper":0,"mentioned_in_github":1,"framework":"tf","reach":{"status":"unanswered"}},{"paper_slug":"pacgan-the-power-of-two-samples-in-generative","repo_url":"https://github.com/xwshen51/AGE","is_official":0,"mentioned_in_paper":0,"mentioned_in_github":1,"framework":"pytorch","reach":{"status":"unanswered"}},{"paper_slug":"pacgan-the-power-of-two-samples-in-generative","repo_url":"https://github.com/xwshen51/AGES","is_official":0,"mentioned_in_paper":0,"mentioned_in_github":1,"framework":"pytorch","reach":{"status":"unanswered"}}],"tasks":[{"task_slug":"diversity","task_name":"Diversity"},{"task_slug":"hypothesis-testing","task_name":"Two-sample testing"},{"task_slug":"two","task_name":"Vocal Bursts Valence Prediction"}],"methods":[],"datasets_introduced":[],"methods_introduced":[],"results":[],"syntology":{"atlas_url":"https://app.syntology.ai/?focus=1712.04086","mcp":null,"developers":"https://syntology.ai/developers"},"arxiv_metadata":null,"syntology_extracted_results":null}