{"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/analysis-and-optimization-of-loss-functions","title":"Analysis and Optimization of Loss Functions for Multiclass, Top-k, and Multilabel Classification","arxiv_id":"1612.03663","date":"2016-12-12","proceeding":null,"authors":["Maksim Lapin","Matthias Hein","Bernt Schiele"],"abstract":"Top-k error is currently a popular performance measure on large scale image\nclassification benchmarks such as ImageNet and Places. Despite its wide\nacceptance, our understanding of this metric is limited as most of the previous\nresearch is focused on its special case, the top-1 error. In this work, we\nexplore two directions that shed more light on the top-k error. First, we\nprovide an in-depth analysis of established and recently proposed single-label\nmulticlass methods along with a detailed account of efficient optimization\nalgorithms for them. Our results indicate that the softmax loss and the smooth\nmulticlass SVM are surprisingly competitive in top-k error uniformly across all\nk, which can be explained by our analysis of multiclass top-k calibration.\nFurther improvements for a specific k are possible with a number of proposed\ntop-k loss functions. Second, we use the top-k methods to explore the\ntransition from multiclass to multilabel learning. In particular, we find that\nit is possible to obtain effective multilabel classifiers on Pascal VOC using a\nsingle label per image for training, while the gap between multiclass and\nmultilabel methods on MS COCO is more significant. Finally, our contribution of\nefficient algorithms for training with the considered top-k and multilabel loss\nfunctions is of independent interest.","url_abs":"http://arxiv.org/abs/1612.03663v1","url_pdf":"http://arxiv.org/pdf/1612.03663v1.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":"analysis-and-optimization-of-loss-functions","repo_url":"https://github.com/mlapin/libsdca","is_official":1,"mentioned_in_paper":1,"mentioned_in_github":0,"framework":"none","reach":{"status":"ok","spdx":"MIT"}}],"tasks":[{"task_slug":"classification","task_name":"General Classification"},{"task_slug":"image-classification","task_name":"Image Classification"},{"task_slug":"image-classification","task_name":"image-classification"}],"methods":[{"method_slug":"svm","method_name":"SVM"},{"method_slug":"softmax","method_name":"Softmax"}],"datasets_introduced":[],"methods_introduced":[],"results":[],"syntology":{"syntology_url":null,"atlas_url":"https://app.syntology.ai/?focus=1612.03663","mcp":null,"developers":"https://syntology.ai/developers"},"arxiv_metadata":null,"syntology_extracted_results":null}