{"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/no-fuss-distance-metric-learning-using","title":"No Fuss Distance Metric Learning using Proxies","arxiv_id":"1703.07464","date":"2017-03-21","proceeding":"ICCV 2017 10","authors":["Yair Movshovitz-Attias","Alexander Toshev","Thomas K. Leung","Sergey Ioffe","Saurabh Singh"],"abstract":"We address the problem of distance metric learning (DML), defined as learning\na distance consistent with a notion of semantic similarity. Traditionally, for\nthis problem supervision is expressed in the form of sets of points that follow\nan ordinal relationship -- an anchor point $x$ is similar to a set of positive\npoints $Y$, and dissimilar to a set of negative points $Z$, and a loss defined\nover these distances is minimized. While the specifics of the optimization\ndiffer, in this work we collectively call this type of supervision Triplets and\nall methods that follow this pattern Triplet-Based methods. These methods are\nchallenging to optimize. A main issue is the need for finding informative\ntriplets, which is usually achieved by a variety of tricks such as increasing\nthe batch size, hard or semi-hard triplet mining, etc. Even with these tricks,\nthe convergence rate of such methods is slow. In this paper we propose to\noptimize the triplet loss on a different space of triplets, consisting of an\nanchor data point and similar and dissimilar proxy points which are learned as\nwell. These proxies approximate the original data points, so that a triplet\nloss over the proxies is a tight upper bound of the original loss. This\nproxy-based loss is empirically better behaved. As a result, the proxy-loss\nimproves on state-of-art results for three standard zero-shot learning\ndatasets, by up to 15% points, while converging three times as fast as other\ntriplet-based losses.","url_abs":"http://arxiv.org/abs/1703.07464v3","url_pdf":"http://arxiv.org/pdf/1703.07464v3.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":"no-fuss-distance-metric-learning-using","repo_url":"https://github.com/QuocThangNguyen/deep-metric-learning-tsinghua-dogs","is_official":0,"mentioned_in_paper":0,"mentioned_in_github":1,"framework":"pytorch","reach":null},{"paper_slug":"no-fuss-distance-metric-learning-using","repo_url":"https://github.com/dichotomies/proxy-nca","is_official":0,"mentioned_in_paper":0,"mentioned_in_github":1,"framework":"pytorch","reach":null}],"tasks":[{"task_slug":"metric-learning","task_name":"Metric Learning"},{"task_slug":"semantic-similarity","task_name":"Semantic Similarity"},{"task_slug":"semantic-textual-similarity","task_name":"Semantic Textual Similarity"},{"task_slug":null,"task_name":"Triplet"},{"task_slug":"zero-shot-learning","task_name":"Zero-Shot Learning"}],"methods":[],"datasets_introduced":[],"methods_introduced":[],"results":[],"syntology":{"syntology_url":"https://syntology.ai/paper/1703.07464","atlas_url":"https://app.syntology.ai/?focus=1703.07464","mcp":null,"developers":"https://syntology.ai/developers"},"arxiv_metadata":null,"syntology_extracted_results":null}