{"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/asymmetric-tri-training-for-unsupervised","title":"Asymmetric Tri-training for Unsupervised Domain Adaptation","arxiv_id":"1702.08400","date":"2017-02-27","proceeding":"ICML 2017 8","authors":["Kuniaki Saito","Yoshitaka Ushiku","Tatsuya Harada"],"abstract":"Deep-layered models trained on a large number of labeled samples boost the\naccuracy of many tasks. It is important to apply such models to different\ndomains because collecting many labeled samples in various domains is\nexpensive. In unsupervised domain adaptation, one needs to train a classifier\nthat works well on a target domain when provided with labeled source samples\nand unlabeled target samples. Although many methods aim to match the\ndistributions of source and target samples, simply matching the distribution\ncannot ensure accuracy on the target domain. To learn discriminative\nrepresentations for the target domain, we assume that artificially labeling\ntarget samples can result in a good representation. Tri-training leverages\nthree classifiers equally to give pseudo-labels to unlabeled samples, but the\nmethod does not assume labeling samples generated from a different domain.In\nthis paper, we propose an asymmetric tri-training method for unsupervised\ndomain adaptation, where we assign pseudo-labels to unlabeled samples and train\nneural networks as if they are true labels. In our work, we use three networks\nasymmetrically. By asymmetric, we mean that two networks are used to label\nunlabeled target samples and one network is trained by the samples to obtain\ntarget-discriminative representations. We evaluate our method on digit\nrecognition and sentiment analysis datasets. Our proposed method achieves\nstate-of-the-art performance on the benchmark digit recognition datasets of\ndomain adaptation.","url_abs":"http://arxiv.org/abs/1702.08400v3","url_pdf":"http://arxiv.org/pdf/1702.08400v3.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":"asymmetric-tri-training-for-unsupervised","repo_url":"https://github.com/ksaito-ut/atda","is_official":0,"mentioned_in_paper":0,"mentioned_in_github":1,"framework":"tf","reach":{"status":"ok"}}],"tasks":[{"task_slug":"domain-adaptation","task_name":"Domain Adaptation"},{"task_slug":"sentiment-analysis","task_name":"Sentiment Analysis"},{"task_slug":"unsupervised-domain-adaptation","task_name":"Unsupervised Domain Adaptation"}],"methods":[],"datasets_introduced":[],"methods_introduced":[],"results":[{"leaderboard":"/sota/sentiment-analysis-on-multi-domain-sentiment","task":"Sentiment Analysis","dataset":"Multi-Domain Sentiment Dataset","model":"Asymmetric tri-training","rank_in_archive_order":5,"of":6,"metrics":{"Average":"78.39","Books":"72.97","DVD":"76.17","Electronics":"80.47","Kitchen":"83.97"},"uses_additional_data":false}],"syntology":{"atlas_url":"https://app.syntology.ai/?focus=1702.08400","mcp":null,"developers":"https://syntology.ai/developers"},"arxiv_metadata":null,"syntology_extracted_results":null}