{"corpus_id":4357800,"paper_sha":"345afa0e85cb2f5cb438ae44027499ff2c392409","doi":"10.1109/CVPR.2017.316","arxiv_id":"1702.05464","pmid":null,"pmcid":null,"mag_id":2902589289,"dblp_id":"journals/corr/TzengHSD17","acl_id":null,"title":"Adversarial Discriminative Domain Adaptation","year":2017,"publication_date":"2017-02-17","venue":"Computer Vision and Pattern Recognition","journal":{"name":"2017 IEEE Conference on Computer Vision and Pattern Recognition (CVPR)","pages":"2962-2971","volume":null},"journal_issn":null,"journal_title":null,"publication_types":["JournalArticle","Conference"],"pubmed_pub_types":null,"s2_fields_of_study":["Computer Science"],"reference_count":32,"citation_count":5118,"influential_citation_count":540,"is_open_access":true,"arxiv_categories":["cs.CV"],"arxiv_license":"http://arxiv.org/licenses/nonexclusive-distrib/1.0/","arxiv_journal_ref":null,"mesh_headings":null,"chemicals":null,"comments_corrections":null,"source_flags":1,"s2_open_access_pdf_url":"https://arxiv.org/pdf/1702.05464","s2_open_access_landing_url":"https://www.semanticscholar.org/paper/345afa0e85cb2f5cb438ae44027499ff2c392409","s2_open_access_license":null,"s2_open_access_status":"GREEN","pmc_open_access_pdf_url":null,"pmc_open_access_landing_url":null,"pmc_open_access_license":null,"pmc_open_access_status":null,"unpaywall_open_access_pdf_url":null,"unpaywall_open_access_landing_url":null,"unpaywall_open_access_license":null,"unpaywall_open_access_status":null,"abstract":"Adversarial learning methods are a promising approach to training robust deep networks, and can generate complex samples across diverse domains. They can also improve recognition despite the presence of domain shift or dataset bias: recent adversarial approaches to unsupervised domain adaptation reduce the difference between the training and test domain distributions and thus improve generalization performance. However, while generative adversarial networks (GANs) show compelling visualizations, they are not optimal on discriminative tasks and can be limited to smaller shifts. On the other hand, discriminative approaches can handle larger domain shifts, but impose tied weights on the model and do not exploit a GAN-based loss. In this work, we first outline a novel generalized framework for adversarial adaptation, which subsumes recent state-of-the-art approaches as special cases, and use this generalized view to better relate prior approaches. We then propose a previously unexplored instance of our general framework which combines discriminative modeling, untied weight sharing, and a GAN loss, which we call Adversarial Discriminative Domain Adaptation (ADDA). We show that ADDA is more effective yet considerably simpler than competing domain-adversarial methods, and demonstrate the promise of our approach by exceeding state-of-the-art unsupervised adaptation results on standard domain adaptation tasks as well as a difficult cross-modality object classification task.","claims":[{"public_id":"cl_db5948f124d3b2703005a1f8da37f08d","status":"active","text":"Adversarial Discriminative Domain Adaptation (ADDA) combines discriminative modeling, untied weight sharing, and a GAN loss, and exceeds state-of-the-art unsupervised adaptation results on standard domain adaptation tasks as well as a difficult cross-modality object classification task.","confidence":0.95,"contributors":[{"id":2,"public_id":"4715169a40","public_label":"AK (4715169a40)","roles":["extraction"],"url":"https://sah.borca.ai/u/4715169a40"},{"id":1,"public_id":"12632b8b5f","public_label":"Anonymous (12632b8b5f)","roles":["review"],"url":"https://sah.borca.ai/u/12632b8b5f"},{"id":136,"public_id":"3c2apqe3ut","public_label":"Anonymous (3c2apqe3ut)","roles":["review"],"url":"https://sah.borca.ai/u/3c2apqe3ut"},{"id":35,"public_id":"b2adb6bfad","public_label":"Anonymous (b2adb6bfad)","roles":["review"],"url":"https://sah.borca.ai/u/b2adb6bfad"}],"url":"https://sah.borca.ai/claims/cl_db5948f124d3b2703005a1f8da37f08d"},{"public_id":"cl_487a797b83dac06a5e594ad69a81c080","status":"active","text":"Generative adversarial networks (GANs) are not optimal on discriminative tasks and can be limited to smaller domain shifts, while discriminative approaches can handle larger domain shifts but impose tied weights and do not exploit a GAN-based loss.","confidence":0.85,"contributors":[{"id":2,"public_id":"4715169a40","public_label":"AK (4715169a40)","roles":["extraction"],"url":"https://sah.borca.ai/u/4715169a40"},{"id":1,"public_id":"12632b8b5f","public_label":"Anonymous (12632b8b5f)","roles":["review"],"url":"https://sah.borca.ai/u/12632b8b5f"},{"id":136,"public_id":"3c2apqe3ut","public_label":"Anonymous (3c2apqe3ut)","roles":["review"],"url":"https://sah.borca.ai/u/3c2apqe3ut"},{"id":35,"public_id":"b2adb6bfad","public_label":"Anonymous (b2adb6bfad)","roles":["review"],"url":"https://sah.borca.ai/u/b2adb6bfad"}],"url":"https://sah.borca.ai/claims/cl_487a797b83dac06a5e594ad69a81c080"}],"concepts":[{"public_id":"co_305996f71d8a2192aa15370d1d4f467e","status":"active","name":"generalized framework for adversarial adaptation","description":"A novel framework that subsumes recent state-of-the-art adversarial adaptation approaches as special cases.","types":["framework"],"aliases":[],"contributors":[{"id":2,"public_id":"4715169a40","public_label":"AK (4715169a40)","roles":["extraction"],"url":"https://sah.borca.ai/u/4715169a40"},{"id":1,"public_id":"12632b8b5f","public_label":"Anonymous (12632b8b5f)","roles":["review"],"url":"https://sah.borca.ai/u/12632b8b5f"},{"id":136,"public_id":"3c2apqe3ut","public_label":"Anonymous (3c2apqe3ut)","roles":["review"],"url":"https://sah.borca.ai/u/3c2apqe3ut"},{"id":35,"public_id":"b2adb6bfad","public_label":"Anonymous (b2adb6bfad)","roles":["review"],"url":"https://sah.borca.ai/u/b2adb6bfad"}],"url":"https://sah.borca.ai/concepts/co_305996f71d8a2192aa15370d1d4f467e"},{"public_id":"co_986c0ebdc1bbd287e7454438bbc96ac8","status":"active","name":"generative adversarial networks","description":"Adversarial learning methods that generate complex samples across diverse domains, but are not optimal on discriminative tasks.","types":["method"],"aliases":["GANs"],"contributors":[{"id":2,"public_id":"4715169a40","public_label":"AK (4715169a40)","roles":["extraction"],"url":"https://sah.borca.ai/u/4715169a40"},{"id":1,"public_id":"12632b8b5f","public_label":"Anonymous (12632b8b5f)","roles":["review"],"url":"https://sah.borca.ai/u/12632b8b5f"},{"id":136,"public_id":"3c2apqe3ut","public_label":"Anonymous (3c2apqe3ut)","roles":["review"],"url":"https://sah.borca.ai/u/3c2apqe3ut"},{"id":35,"public_id":"b2adb6bfad","public_label":"Anonymous (b2adb6bfad)","roles":["review"],"url":"https://sah.borca.ai/u/b2adb6bfad"}],"url":"https://sah.borca.ai/concepts/co_986c0ebdc1bbd287e7454438bbc96ac8"},{"public_id":"co_a0a1ca62ff247f1253d7b1fad03ef34d","status":"active","name":"untied weight sharing","description":"Weight sharing strategy where weights are not tied between the source and target models.","types":["technique"],"aliases":[],"contributors":[{"id":2,"public_id":"4715169a40","public_label":"AK (4715169a40)","roles":["extraction"],"url":"https://sah.borca.ai/u/4715169a40"},{"id":1,"public_id":"12632b8b5f","public_label":"Anonymous (12632b8b5f)","roles":["review"],"url":"https://sah.borca.ai/u/12632b8b5f"},{"id":136,"public_id":"3c2apqe3ut","public_label":"Anonymous (3c2apqe3ut)","roles":["review"],"url":"https://sah.borca.ai/u/3c2apqe3ut"},{"id":35,"public_id":"b2adb6bfad","public_label":"Anonymous (b2adb6bfad)","roles":["review"],"url":"https://sah.borca.ai/u/b2adb6bfad"}],"url":"https://sah.borca.ai/concepts/co_a0a1ca62ff247f1253d7b1fad03ef34d"},{"public_id":"co_d303474686182d11d647ed01cca4b026","status":"active","name":"unsupervised domain adaptation","description":"Domain adaptation setting where labeled data is available only in the source domain, not the target domain.","types":["task setting"],"aliases":[],"contributors":[{"id":2,"public_id":"4715169a40","public_label":"AK (4715169a40)","roles":["extraction"],"url":"https://sah.borca.ai/u/4715169a40"},{"id":1,"public_id":"12632b8b5f","public_label":"Anonymous (12632b8b5f)","roles":["review"],"url":"https://sah.borca.ai/u/12632b8b5f"},{"id":136,"public_id":"3c2apqe3ut","public_label":"Anonymous (3c2apqe3ut)","roles":["review"],"url":"https://sah.borca.ai/u/3c2apqe3ut"},{"id":35,"public_id":"b2adb6bfad","public_label":"Anonymous (b2adb6bfad)","roles":["review"],"url":"https://sah.borca.ai/u/b2adb6bfad"}],"url":"https://sah.borca.ai/concepts/co_d303474686182d11d647ed01cca4b026"},{"public_id":"co_db7a1ae3319aaa411947779be1c8fb1c","status":"active","name":"discriminative approaches","description":"Approaches that handle larger domain shifts but impose tied weights on the model and do not exploit a GAN-based loss.","types":["approach"],"aliases":[],"contributors":[{"id":2,"public_id":"4715169a40","public_label":"AK (4715169a40)","roles":["extraction"],"url":"https://sah.borca.ai/u/4715169a40"},{"id":1,"public_id":"12632b8b5f","public_label":"Anonymous (12632b8b5f)","roles":["review"],"url":"https://sah.borca.ai/u/12632b8b5f"},{"id":136,"public_id":"3c2apqe3ut","public_label":"Anonymous (3c2apqe3ut)","roles":["review"],"url":"https://sah.borca.ai/u/3c2apqe3ut"},{"id":35,"public_id":"b2adb6bfad","public_label":"Anonymous (b2adb6bfad)","roles":["review"],"url":"https://sah.borca.ai/u/b2adb6bfad"}],"url":"https://sah.borca.ai/concepts/co_db7a1ae3319aaa411947779be1c8fb1c"},{"public_id":"co_e526e6b09c34ffe741920722c6b10e40","status":"active","name":"discriminative modeling","description":"Modeling approach focused on discriminative tasks rather than generative sample generation.","types":["approach"],"aliases":[],"contributors":[{"id":2,"public_id":"4715169a40","public_label":"AK (4715169a40)","roles":["extraction"],"url":"https://sah.borca.ai/u/4715169a40"},{"id":1,"public_id":"12632b8b5f","public_label":"Anonymous (12632b8b5f)","roles":["review"],"url":"https://sah.borca.ai/u/12632b8b5f"},{"id":136,"public_id":"3c2apqe3ut","public_label":"Anonymous (3c2apqe3ut)","roles":["review"],"url":"https://sah.borca.ai/u/3c2apqe3ut"},{"id":35,"public_id":"b2adb6bfad","public_label":"Anonymous (b2adb6bfad)","roles":["review"],"url":"https://sah.borca.ai/u/b2adb6bfad"}],"url":"https://sah.borca.ai/concepts/co_e526e6b09c34ffe741920722c6b10e40"},{"public_id":"co_f4e3d16098071659996934185bba86fd","status":"active","name":"Adversarial Discriminative Domain Adaptation","description":"A proposed instance of the general adversarial adaptation framework that combines discriminative modeling, untied weight sharing, and a GAN loss.","types":["method"],"aliases":["ADDA"],"contributors":[{"id":2,"public_id":"4715169a40","public_label":"AK (4715169a40)","roles":["extraction"],"url":"https://sah.borca.ai/u/4715169a40"},{"id":1,"public_id":"12632b8b5f","public_label":"Anonymous (12632b8b5f)","roles":["review"],"url":"https://sah.borca.ai/u/12632b8b5f"},{"id":136,"public_id":"3c2apqe3ut","public_label":"Anonymous (3c2apqe3ut)","roles":["review"],"url":"https://sah.borca.ai/u/3c2apqe3ut"},{"id":35,"public_id":"b2adb6bfad","public_label":"Anonymous (b2adb6bfad)","roles":["review"],"url":"https://sah.borca.ai/u/b2adb6bfad"}],"url":"https://sah.borca.ai/concepts/co_f4e3d16098071659996934185bba86fd"},{"public_id":"co_fb76277c2b0ac5acc55b627384e390c1","status":"active","name":"GAN loss","description":"Loss function derived from generative adversarial networks used for adversarial adaptation.","types":["loss function"],"aliases":[],"contributors":[{"id":2,"public_id":"4715169a40","public_label":"AK (4715169a40)","roles":["extraction"],"url":"https://sah.borca.ai/u/4715169a40"},{"id":1,"public_id":"12632b8b5f","public_label":"Anonymous (12632b8b5f)","roles":["review"],"url":"https://sah.borca.ai/u/12632b8b5f"},{"id":136,"public_id":"3c2apqe3ut","public_label":"Anonymous (3c2apqe3ut)","roles":["review"],"url":"https://sah.borca.ai/u/3c2apqe3ut"},{"id":35,"public_id":"b2adb6bfad","public_label":"Anonymous (b2adb6bfad)","roles":["review"],"url":"https://sah.borca.ai/u/b2adb6bfad"}],"url":"https://sah.borca.ai/concepts/co_fb76277c2b0ac5acc55b627384e390c1"}],"external_ids":{"DOI":"10.1109/CVPR.2017.316","ArXiv":"1702.05464","PubMed":null,"PubMedCentral":null,"MAG":2902589289,"DBLP":"journals/corr/TzengHSD17","ACL":null},"open_access":{"is_open_access":true,"pdf_url":"https://arxiv.org/pdf/1702.05464","landing_url":"https://www.semanticscholar.org/paper/345afa0e85cb2f5cb438ae44027499ff2c392409","source":"semantic_scholar","pdf_url_source":"semantic_scholar_open_access_pdf","license":null,"status":"GREEN","reason":null},"reference_availability":{"status":"available","references_indexed":true,"full_text_available":true,"full_text_source":"arxiv","count_basis":"semantic_scholar_metadata","extraction_status":"not_applicable","reason":null},"source":{"provider":"episteme2","base_corpus":"semantic_scholar_dump","freshness_mode":"unknown","basis":["semantic_scholar_metadata","postgres_metadata"],"limits":["paper metadata is based on indexed upstream scholarly datasets","claims and concepts are available only for extracted papers","absence of claims or concepts means no extracted graph data is available in this response"],"status":"available","degraded":false,"degraded_reasons":[],"diagnostics":{"status":"available","degraded":false,"degraded_reasons":[],"metadata_status":"available","graph_status":"available","abstract_status":"available"},"source_flags":1},"paper_id":633037,"paper_uid":"59a40f7c-a259-40dd-8d00-423d862923a0","canonical_identity":{"paper_id":633037,"paper_uid":"59a40f7c-a259-40dd-8d00-423d862923a0","identity_status":"available","lookup_basis":"semantic_scholar_external_id","compatibility_path":"corpus_id"},"url":"https://sah.borca.ai/papers/4357800"}