{"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/a-dirt-t-approach-to-unsupervised-domain","title":"A DIRT-T Approach to Unsupervised Domain Adaptation","arxiv_id":"1802.08735","date":"2018-02-23","proceeding":"ICLR 2018 1","authors":["Rui Shu","Hung H. Bui","Hirokazu Narui","Stefano Ermon"],"abstract":"Domain adaptation refers to the problem of leveraging labeled data in a\nsource domain to learn an accurate model in a target domain where labels are\nscarce or unavailable. A recent approach for finding a common representation of\nthe two domains is via domain adversarial training (Ganin & Lempitsky, 2015),\nwhich attempts to induce a feature extractor that matches the source and target\nfeature distributions in some feature space. However, domain adversarial\ntraining faces two critical limitations: 1) if the feature extraction function\nhas high-capacity, then feature distribution matching is a weak constraint, 2)\nin non-conservative domain adaptation (where no single classifier can perform\nwell in both the source and target domains), training the model to do well on\nthe source domain hurts performance on the target domain. In this paper, we\naddress these issues through the lens of the cluster assumption, i.e., decision\nboundaries should not cross high-density data regions. We propose two novel and\nrelated models: 1) the Virtual Adversarial Domain Adaptation (VADA) model,\nwhich combines domain adversarial training with a penalty term that punishes\nthe violation the cluster assumption; 2) the Decision-boundary Iterative\nRefinement Training with a Teacher (DIRT-T) model, which takes the VADA model\nas initialization and employs natural gradient steps to further minimize the\ncluster assumption violation. Extensive empirical results demonstrate that the\ncombination of these two models significantly improve the state-of-the-art\nperformance on the digit, traffic sign, and Wi-Fi recognition domain adaptation\nbenchmarks.","url_abs":"http://arxiv.org/abs/1802.08735v2","url_pdf":"http://arxiv.org/pdf/1802.08735v2.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":"a-dirt-t-approach-to-unsupervised-domain","repo_url":"https://github.com/RuiShu/dirt-t","is_official":1,"mentioned_in_paper":1,"mentioned_in_github":1,"framework":"tf","reach":{"status":"ok","spdx":"MIT"}},{"paper_slug":"a-dirt-t-approach-to-unsupervised-domain","repo_url":"https://github.com/domainadaptation/salad","is_official":0,"mentioned_in_paper":0,"mentioned_in_github":1,"framework":"pytorch","reach":{"status":"ok","spdx":"MPL-2.0"}},{"paper_slug":"a-dirt-t-approach-to-unsupervised-domain","repo_url":"https://github.com/timgaripov/asa","is_official":0,"mentioned_in_paper":0,"mentioned_in_github":1,"framework":"pytorch","reach":null},{"paper_slug":"a-dirt-t-approach-to-unsupervised-domain","repo_url":"https://github.com/kevinmusgrave/pytorch-adapt","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":"unsupervised-domain-adaptation","task_name":"Unsupervised Domain Adaptation"}],"methods":[],"datasets_introduced":[],"methods_introduced":[],"results":[],"syntology":{"syntology_url":"https://syntology.ai/paper/1802.08735","atlas_url":"https://app.syntology.ai/?focus=1802.08735","mcp":null,"developers":"https://syntology.ai/developers"},"arxiv_metadata":null,"syntology_extracted_results":null}