{"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/learning-to-cluster-in-order-to-transfer","title":"Learning to cluster in order to transfer across domains and tasks","arxiv_id":"1711.10125","date":"2017-11-28","proceeding":"ICLR 2018 1","authors":["Yen-Chang Hsu","Zhaoyang Lv","Zsolt Kira"],"abstract":"This paper introduces a novel method to perform transfer learning across\ndomains and tasks, formulating it as a problem of learning to cluster. The key\ninsight is that, in addition to features, we can transfer similarity\ninformation and this is sufficient to learn a similarity function and\nclustering network to perform both domain adaptation and cross-task transfer\nlearning. We begin by reducing categorical information to pairwise constraints,\nwhich only considers whether two instances belong to the same class or not.\nThis similarity is category-agnostic and can be learned from data in the source\ndomain using a similarity network. We then present two novel approaches for\nperforming transfer learning using this similarity function. First, for\nunsupervised domain adaptation, we design a new loss function to regularize\nclassification with a constrained clustering loss, hence learning a clustering\nnetwork with the transferred similarity metric generating the training inputs.\nSecond, for cross-task learning (i.e., unsupervised clustering with unseen\ncategories), we propose a framework to reconstruct and estimate the number of\nsemantic clusters, again using the clustering network. Since the similarity\nnetwork is noisy, the key is to use a robust clustering algorithm, and we show\nthat our formulation is more robust than the alternative constrained and\nunconstrained clustering approaches. Using this method, we first show state of\nthe art results for the challenging cross-task problem, applied on Omniglot and\nImageNet. Our results show that we can reconstruct semantic clusters with high\naccuracy. We then evaluate the performance of cross-domain transfer using\nimages from the Office-31 and SVHN-MNIST tasks and present top accuracy on both\ndatasets. Our approach doesn't explicitly deal with domain discrepancy. If we\ncombine with a domain adaptation loss, it shows further improvement.","url_abs":"http://arxiv.org/abs/1711.10125v3","url_pdf":"http://arxiv.org/pdf/1711.10125v3.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":"learning-to-cluster-in-order-to-transfer","repo_url":"https://github.com/GT-RIPL/L2C","is_official":0,"mentioned_in_paper":0,"mentioned_in_github":1,"framework":"pytorch","reach":{"status":"ok","spdx":"MIT"}}],"tasks":[{"task_slug":"clustering","task_name":"Clustering"},{"task_slug":"constrained-clustering","task_name":"Constrained Clustering"},{"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":[],"syntology":{"syntology_url":"https://syntology.ai/paper/1711.10125","atlas_url":"https://app.syntology.ai/?focus=1711.10125","mcp":null,"developers":"https://syntology.ai/developers"},"arxiv_metadata":null,"syntology_extracted_results":null}