{"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/visual-domain-adaptation-with-manifold","title":"Visual Domain Adaptation with Manifold Embedded Distribution Alignment","arxiv_id":"1807.07258","date":"2018-07-19","proceeding":null,"authors":["Jindong Wang","Wenjie Feng","Yiqiang Chen","Han Yu","Meiyu Huang","Philip S. Yu"],"abstract":"Visual domain adaptation aims to learn robust classifiers for the target\ndomain by leveraging knowledge from a source domain. Existing methods either\nattempt to align the cross-domain distributions, or perform manifold subspace\nlearning. However, there are two significant challenges: (1) degenerated\nfeature transformation, which means that distribution alignment is often\nperformed in the original feature space, where feature distortions are hard to\novercome. On the other hand, subspace learning is not sufficient to reduce the\ndistribution divergence. (2) unevaluated distribution alignment, which means\nthat existing distribution alignment methods only align the marginal and\nconditional distributions with equal importance, while they fail to evaluate\nthe different importance of these two distributions in real applications. In\nthis paper, we propose a Manifold Embedded Distribution Alignment (MEDA)\napproach to address these challenges. MEDA learns a domain-invariant classifier\nin Grassmann manifold with structural risk minimization, while performing\ndynamic distribution alignment to quantitatively account for the relative\nimportance of marginal and conditional distributions. To the best of our\nknowledge, MEDA is the first attempt to perform dynamic distribution alignment\nfor manifold domain adaptation. Extensive experiments demonstrate that MEDA\nshows significant improvements in classification accuracy compared to\nstate-of-the-art traditional and deep methods.","url_abs":"http://arxiv.org/abs/1807.07258v2","url_pdf":"http://arxiv.org/pdf/1807.07258v2.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":"visual-domain-adaptation-with-manifold","repo_url":"https://github.com/jindongwang/transferlearning","is_official":0,"mentioned_in_paper":0,"mentioned_in_github":1,"framework":"pytorch","reach":{"status":"ok","spdx":"MIT"}}],"tasks":[{"task_slug":"domain-adaptation","task_name":"Domain Adaptation"},{"task_slug":"transfer-learning","task_name":"Transfer Learning"},{"task_slug":"unsupervised-domain-adaptation","task_name":"Unsupervised Domain Adaptation"}],"methods":[],"datasets_introduced":[],"methods_introduced":[],"results":[{"leaderboard":"/sota/domain-adaptation-on-office-caltech","task":"Domain Adaptation","dataset":"Office-Caltech","model":"MEDA[[Wang et al.2018]]","rank_in_archive_order":2,"of":8,"metrics":{"Average Accuracy":"92.8"},"uses_additional_data":false},{"leaderboard":"/sota/domain-adaptation-on-office-caltech-10","task":"Domain Adaptation","dataset":"Office-Caltech-10","model":"MEDA","rank_in_archive_order":1,"of":1,"metrics":{"Accuracy (%)":"92.8"},"uses_additional_data":false},{"leaderboard":"/sota/transfer-learning-on-office-home","task":"Transfer Learning","dataset":"Office-Home","model":"MEDA","rank_in_archive_order":5,"of":5,"metrics":{"Accuracy":"60.3"},"uses_additional_data":false}],"syntology":{"syntology_url":"https://syntology.ai/paper/1807.07258","atlas_url":"https://app.syntology.ai/?focus=1807.07258","mcp":null,"developers":"https://syntology.ai/developers"},"arxiv_metadata":null,"syntology_extracted_results":null}