{"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/on-line-adaptative-curriculum-learning-for","title":"On-line Adaptative Curriculum Learning for GANs","arxiv_id":"1808.00020","date":"2018-07-31","proceeding":null,"authors":["Thang Doan","Joao Monteiro","Isabela Albuquerque","Bogdan Mazoure","Audrey Durand","Joelle Pineau","R. Devon Hjelm"],"abstract":"Generative Adversarial Networks (GANs) can successfully approximate a\nprobability distribution and produce realistic samples. However, open questions\nsuch as sufficient convergence conditions and mode collapse still persist. In\nthis paper, we build on existing work in the area by proposing a novel\nframework for training the generator against an ensemble of discriminator\nnetworks, which can be seen as a one-student/multiple-teachers setting. We\nformalize this problem within the full-information adversarial bandit\nframework, where we evaluate the capability of an algorithm to select mixtures\nof discriminators for providing the generator with feedback during learning. To\nthis end, we propose a reward function which reflects the progress made by the\ngenerator and dynamically update the mixture weights allocated to each\ndiscriminator. We also draw connections between our algorithm and stochastic\noptimization methods and then show that existing approaches using multiple\ndiscriminators in literature can be recovered from our framework. We argue that\nless expressive discriminators are smoother and have a general coarse grained\nview of the modes map, which enforces the generator to cover a wide portion of\nthe data distribution support. On the other hand, highly expressive\ndiscriminators ensure samples quality. Finally, experimental results show that\nour approach improves samples quality and diversity over existing baselines by\neffectively learning a curriculum. These results also support the claim that\nweaker discriminators have higher entropy improving modes coverage. Keywords:\nmultiple discriminators, curriculum learning, multiple resolutions\ndiscriminators, multi-armed bandits, generative adversarial networks, smooth\ndiscriminators, multi-discriminator gan training, multiple experts.","url_abs":"http://arxiv.org/abs/1808.00020v6","url_pdf":"http://arxiv.org/pdf/1808.00020v6.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":"on-line-adaptative-curriculum-learning-for","repo_url":"https://github.com/Byte7/Adaptative-Curriculum-GAN-keras","is_official":0,"mentioned_in_paper":0,"mentioned_in_github":1,"framework":"none","reach":{"status":"ok"}},{"paper_slug":"on-line-adaptative-curriculum-learning-for","repo_url":"https://github.com/Byte7/Adaptive-Curriculum-GAN-keras","is_official":0,"mentioned_in_paper":0,"mentioned_in_github":1,"framework":"none","reach":{"status":"ok"}},{"paper_slug":"on-line-adaptative-curriculum-learning-for","repo_url":"https://github.com/Sirius79/acGAN","is_official":0,"mentioned_in_paper":0,"mentioned_in_github":1,"framework":"pytorch","reach":{"status":"ok"}}],"tasks":[{"task_slug":"multi-armed-bandits","task_name":"Multi-Armed Bandits"},{"task_slug":"stochastic-optimization","task_name":"Stochastic Optimization"}],"methods":[{"method_slug":"convolution","method_name":"Convolution"}],"datasets_introduced":[],"methods_introduced":[],"results":[],"syntology":{"syntology_url":"https://syntology.ai/paper/1808.00020","atlas_url":"https://app.syntology.ai/?focus=1808.00020","mcp":null,"developers":"https://syntology.ai/developers"},"arxiv_metadata":null,"syntology_extracted_results":null}