{"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/augmented-cyclic-adversarial-learning-for-low","title":"Augmented Cyclic Adversarial Learning for Low Resource Domain Adaptation","arxiv_id":"1807.00374","date":"2018-07-01","proceeding":"ICLR 2019 5","authors":["Ehsan Hosseini-Asl","Yingbo Zhou","Caiming Xiong","Richard Socher"],"abstract":"Training a model to perform a task typically requires a large amount of data\nfrom the domains in which the task will be applied. However, it is often the\ncase that data are abundant in some domains but scarce in others. Domain\nadaptation deals with the challenge of adapting a model trained from a\ndata-rich source domain to perform well in a data-poor target domain. In\ngeneral, this requires learning plausible mappings between domains. CycleGAN is\na powerful framework that efficiently learns to map inputs from one domain to\nanother using adversarial training and a cycle-consistency constraint. However,\nthe conventional approach of enforcing cycle-consistency via reconstruction may\nbe overly restrictive in cases where one or more domains have limited training\ndata. In this paper, we propose an augmented cyclic adversarial learning model\nthat enforces the cycle-consistency constraint via an external task specific\nmodel, which encourages the preservation of task-relevant content as opposed to\nexact reconstruction. We explore digit classification in a low-resource setting\nin supervised, semi and unsupervised situation, as well as high resource\nunsupervised. In low-resource supervised setting, the results show that our\napproach improves absolute performance by 14% and 4% when adapting SVHN to\nMNIST and vice versa, respectively, which outperforms unsupervised domain\nadaptation methods that require high-resource unlabeled target domain.\nMoreover, using only few unsupervised target data, our approach can still\noutperforms many high-resource unsupervised models. In speech domains, we\nsimilarly adopt a speech recognition model from each domain as the task\nspecific model. Our approach improves absolute performance of speech\nrecognition by 2% for female speakers in the TIMIT dataset, where the majority\nof training samples are from male voices.","url_abs":"http://arxiv.org/abs/1807.00374v4","url_pdf":"http://arxiv.org/pdf/1807.00374v4.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":"augmented-cyclic-adversarial-learning-for-low","repo_url":"https://github.com/kiniavinash/ACAL","is_official":0,"mentioned_in_paper":0,"mentioned_in_github":1,"framework":"pytorch","reach":null},{"paper_slug":"augmented-cyclic-adversarial-learning-for-low","repo_url":"https://github.com/ms903-github/ACAL-pytorch","is_official":0,"mentioned_in_paper":0,"mentioned_in_github":0,"framework":"pytorch","reach":null}],"tasks":[{"task_slug":"domain-adaptation","task_name":"Domain Adaptation"},{"task_slug":"speech-recognition","task_name":"Speech Recognition"},{"task_slug":"unsupervised-domain-adaptation","task_name":"Unsupervised Domain Adaptation"},{"task_slug":"speech-recognition-1","task_name":"speech-recognition"}],"methods":[{"method_slug":"batch-normalization","method_name":"Batch Normalization"},{"method_slug":"convolution","method_name":"Convolution"},{"method_slug":"cycle-consistency-loss","method_name":"Cycle Consistency Loss"},{"method_slug":"gan-least-squares-loss","method_name":"GAN Least Squares Loss"},{"method_slug":"instance-normalization","method_name":"Instance Normalization"},{"method_slug":"patchgan","method_name":"PatchGAN"},{"method_slug":"relu","method_name":"ReLU"},{"method_slug":"residual-block","method_name":"Residual Block"},{"method_slug":"residual-connection","method_name":"Residual Connection"},{"method_slug":"sigmoid-activation","method_name":"Sigmoid Activation"},{"method_slug":"tanh-activation","method_name":"Tanh Activation"}],"datasets_introduced":[],"methods_introduced":[],"results":[],"syntology":{"atlas_url":"https://app.syntology.ai/?focus=1807.00374","mcp":null,"developers":"https://syntology.ai/developers"},"arxiv_metadata":null,"syntology_extracted_results":null}