{"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/minimizing-close-k-aggregate-loss-improves","title":"Minimizing Close-k Aggregate Loss Improves Classification","arxiv_id":"1811.00521","date":"2018-11-01","proceeding":null,"authors":["Bryan He","James Zou"],"abstract":"In classification, the de facto method for aggregating individual losses is\nthe average loss. When the actual metric of interest is 0-1 loss, it is common\nto minimize the average surrogate loss for some well-behaved (e.g. convex)\nsurrogate. Recently, several other aggregate losses such as the maximal loss\nand average top-$k$ loss were proposed as alternative objectives to address\nshortcomings of the average loss. However, we identify common classification\nsettings, e.g. the data is imbalanced, has too many easy or ambiguous examples,\netc., when average, maximal and average top-$k$ all suffer from suboptimal\ndecision boundaries, even on an infinitely large training set. To address this\nproblem, we propose a new classification objective called the close-$k$\naggregate loss, where we adaptively minimize the loss for points close to the\ndecision boundary. We provide theoretical guarantees for the 0-1 accuracy when\nwe optimize close-$k$ aggregate loss. We also conduct systematic experiments\nacross the PMLB and OpenML benchmark datasets. Close-$k$ achieves significant\ngains in 0-1 test accuracy, improvements of $\\geq 2$% and $p<0.05$, in over 25%\nof the datasets compared to average, maximal and average top-$k$. In contrast,\nthe previous aggregate losses outperformed close-$k$ in less than 2% of the\ndatasets.","url_abs":"http://arxiv.org/abs/1811.00521v2","url_pdf":"http://arxiv.org/pdf/1811.00521v2.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":"minimizing-close-k-aggregate-loss-improves","repo_url":"https://github.com/bryan-he/closek","is_official":1,"mentioned_in_paper":1,"mentioned_in_github":1,"framework":"pytorch","reach":null}],"tasks":[{"task_slug":"classification-1","task_name":"Classification"},{"task_slug":"classification","task_name":"General Classification"}],"methods":[],"datasets_introduced":[],"methods_introduced":[],"results":[],"syntology":{"atlas_url":null,"mcp":null,"developers":"https://syntology.ai/developers"},"arxiv_metadata":null,"syntology_extracted_results":null}