[{"data":1,"prerenderedAt":-1},["ShallowReactive",2],{"detail-sidebar-cat-0-en-105":3,"doc-seo-290901-105":59,"doc-detail-290901-en":131},{"code":4,"msg":5,"data":6},0,"success",[7,13,18,23,28,33,38,43,48,51,55],{"id":8,"doc_module":4,"doc_module_name":9,"category_name":10,"show_sort_weight":11,"slug":12},1,"Document","Story & Novel",90,"story-novel",{"id":14,"doc_module":4,"doc_module_name":9,"category_name":15,"show_sort_weight":16,"slug":17},2,"Literature",80,"literature",{"id":19,"doc_module":4,"doc_module_name":9,"category_name":20,"show_sort_weight":21,"slug":22},4,"Exam",70,"exam",{"id":24,"doc_module":4,"doc_module_name":9,"category_name":25,"show_sort_weight":26,"slug":27},5,"Comic",60,"comic",{"id":29,"doc_module":4,"doc_module_name":9,"category_name":30,"show_sort_weight":31,"slug":32},6,"Technology",50,"technology",{"id":34,"doc_module":4,"doc_module_name":9,"category_name":35,"show_sort_weight":36,"slug":37},7,"Healthcare",40,"healthcare",{"id":39,"doc_module":4,"doc_module_name":9,"category_name":40,"show_sort_weight":41,"slug":42},8,"Research & Report",30,"research-report",{"id":44,"doc_module":4,"doc_module_name":9,"category_name":45,"show_sort_weight":46,"slug":47},9,"Religion & Spirituality",20,"religion-spirituality",{"id":46,"doc_module":4,"doc_module_name":9,"category_name":49,"show_sort_weight":46,"slug":50},"World Cup","world-cup",{"id":52,"doc_module":4,"doc_module_name":9,"category_name":53,"show_sort_weight":52,"slug":54},10,"Lifestyle","lifestyle",{"id":56,"doc_module":4,"doc_module_name":9,"category_name":57,"show_sort_weight":24,"slug":58},19,"General","general",{"code":4,"msg":60,"data":61},"ok",{"site_id":62,"language":63,"slug":64,"title":65,"keywords":66,"description":67,"schema_data":68,"social_meta":124,"head_meta":126,"extra_data":128,"updated_unix":130},105,"en","exploratory-machine-learning-with-unknown-unknowns","Exploratory Machine Learning with Unknown Unknowns","","In conventional supervised learning, training data includes ground-truth labels from a known label set, enabling classification of unseen instances among those labels. This paper studies a setting where unknown classes are present in training data but are misperceived as other labels, making their existence invisible under the given supervision. The work attributes this to an incompletely perceived label space caused by insufficient feature information and proposes exploratory machine learning with rejection modeling, feature acquisition, and model cascading, validated on synthetic and real datasets.",{"@graph":69,"@context":123},[70,84,106],{"@type":71,"itemListElement":72},"BreadcrumbList",[73,77,79,82],{"item":74,"name":75,"@type":76,"position":8},"https://docshare.wps.com","Home","ListItem",{"item":78,"name":9,"@type":76,"position":14},"https://docshare.wps.com/document/",{"item":80,"name":40,"@type":76,"position":81},"https://docshare.wps.com/document/research-report/",3,{"item":83,"name":65,"@type":76,"position":19},"https://docshare.wps.com/document/exploratory-machine-learning-with-unknown-unknowns/290901/",{"url":83,"name":65,"@type":85,"image":86,"author":91,"headline":65,"publisher":94,"fileFormat":97,"inLanguage":63,"description":67,"dateModified":98,"datePublished":99,"encodingFormat":97,"isAccessibleForFree":100,"interactionStatistic":101},"DigitalDocument",{"url":87,"@type":88,"width":89,"height":90},"https://docshare.wps.com/thumbnails/exploratory-machine-learning-with-unknown-unknowns/290901.png","ImageObject",300,407,{"name":92,"@type":93},"Eden","Person",{"url":74,"name":95,"@type":96},"DocShare","Organization","application/pdf","2026-10-17","2026-09-17",true,{"@type":102,"interactionType":103,"userInteractionCount":105},"InteractionCounter",{"@type":104},"ViewAction",11,{"@type":107,"mainEntity":108},"FAQPage",[109,115,119],{"name":110,"@type":111,"acceptedAnswer":112},"What problem does the paper call “unknown unknowns” in supervised learning?","Question",{"text":113,"@type":114},"It refers to unknown classes in the training data that are misperceived as other known labels, so their existence is not revealed by the provided supervision.","Answer",{"name":116,"@type":111,"acceptedAnswer":117},"Why do unknown classes emerge in the proposed setting?",{"text":118,"@type":114},"The paper attributes the issue to an incompletely perceived label space caused by insufficient feature information, which makes feature space and label space entangled and unreliable.",{"name":120,"@type":111,"acceptedAnswer":121},"What are the main components of the proposed exploratory learning approach?",{"text":122,"@type":114},"The approach consists of a rejection model, feature acquisition, and a model cascade to actively augment the feature space and discover potentially unknown labels.","https://schema.org",{"og:url":83,"og:type":125,"og:title":65,"og:site_name":95,"og:description":67},"article",{"robots":127,"canonical":83},"index,follow",{"doc_id":129,"site_id":62},290901,1789642922,{"code":4,"msg":5,"data":132},{"doc_id":129,"user_id":133,"nickname":92,"user_avatar":134,"doc_module":4,"category_id":39,"category_name":40,"doc_title":65,"doc_description":67,"doc_content":135,"file_id":136,"file_url":137,"file_type":138,"file_size":139,"view_count":105,"is_deleted":4,"is_public":8,"is_downloadable":8,"audit_status":8,"page_count":140,"language":141,"language_code":63,"site_id":62,"html_lang":63,"table_of_contents":142,"faqs":143,"seo_title":144,"seo_description":67,"update_tm":130,"read_time":145,"preview_page_count":4},1374391974468,"https://ap-avatar.wpscdn.com/davatar_29158cc5080c5b710cf443261637dec0","View metadata, citation and similar [papers at ](papers at core.ac.uk)[core.ac.uk](papers at core.ac.uk) brought to you by CORE  \n[provided by](provided by arXiv.org)[ arXiv.org](provided by arXiv.org) e-Print Archive  \narXiv :2002 .0 1605v 1 [ cs .LG] 5 Feb 2020  \nExploratory Machine Learning with Unknown Unknowns  \nYu-Jie Zhang, Peng Zhao, Zhi-Hua Zhou􀀃  \nNational Key Laboratory for Novel Software Technology  \nNanjing University, Nanjing 210093, China  \nAbstract  \nIn conventional supervised learning, a training dataset is given with ground-truth labels from a known label set, and the learned model will classify unseen instances to the known labels. In this paper, we study a new problem setting in which there are unknown classes in the training dataset misperceived as other labels, and thus their existence appears unknown from the given supervision. We attribute the unknown unknowns to the fact that the training dataset is badly advised by the incompletely perceived label space due to the insu􀀎cient feature information. To this end, we propose the exploratory machine learning, which examines and investigates the training dataset by actively augmenting the feature space to discover potentially unknown labels. Our approach consists of three ingredients including rejection model, feature acquisition, and model cascade. The e􀀋ectiveness is validated on both synthetic and real datasets.  \n1. Introduction  \nIn this paper, we study the task in which there are unknown labels in the training dataset, namely some training instances belonging to a certain class are wrongly perceived as others, and thus appear unknown to the learned model. This is always the case when the label space is misspeci􀀌ed due to the insu􀀎cient feature information. Consider the task of medical diagnosis, where we need to train a model for community healthcare centers based on their patient records, to help diagnose the cause of a patient with cough and dyspnea. As shown in Figure 1, there are actually three causes: two common ones (asthma and pneumonia), as well as an unusual one (lung cancer ) whose diagnosis crucially relies on the computerized tomography (CT) scan device, yet too expensive to purchase. Thus, the community healthcare centers are not likely to diagnose patients with dyspepsia as cancer, resulting in that the class of \\lung cancer\" becomes invisible in the collected training dataset, and thus the learned model will be unaware of this unobserved class.  \n􀀃 Corresponding author. Email: [zhouzh@nju.edu.cn](zhouzh@nju.edu.cn)  \nPreprint submitted for review February 6, 2020  \nground-truth label space  \ncommunity healthcare centers  \nasthma  \npneumonia lung cancer  \nFigure 1: An illustration for the emerging unknown unknowns in the task of medical diagnosis, where the patients with lung cancer are misdiagnosed as asthma or pneumonia due to the lack of CT scan devices, and thus appear as unknown to the learned model.  \nSimilar phenomena occur in many other applications. For instance, the trace of a new-type aircraft was mislabeled as old-type aircrafts until performance of the aircraft detector is found poor (i.e., capability of collected signals is inadequate), and the o􀀎cer suspects that there are new-type aircrafts unknown previously. When feature information is insu􀀎cient, there is a high risk for the learner to misperceive some classes existing in the training data as others, leading to the existence of unknown classes. Especially, unknown classes are sometimes of more interest, like in the above two examples. Therefore, it is crucial for the learned model to discover unknown classes and classify known classes well simultaneously.  \nHowever, the conventional supervised learning (SL), where a predictive model is trained on a given labeled dataset and then deployed to classify unseen instances into known labels, crucially relies on a high-quality training dataset. Thus, when the aforementioned unknown unknowns emerged in the training data, SL canno","cbCaifPnY6a4NVrq","https://ap.wps.com/l/cbCaifPnY6a4NVrq","pdf",991369,21,"English","# Introduction\n# ExML: A New Learning Framework\n## Exploratory Machine Learning","[{\"question\":\"What problem does the paper call “unknown unknowns” in supervised learning?\",\"answer\":\"It refers to unknown classes in the training data that are misperceived as other known labels, so their existence is not revealed by the provided supervision.\"},{\"question\":\"Why do unknown classes emerge in the proposed setting?\",\"answer\":\"The paper attributes the issue to an incompletely perceived label space caused by insufficient feature information, which makes feature space and label space entangled and unreliable.\"},{\"question\":\"What are the main components of the proposed exploratory learning approach?\",\"answer\":\"The approach consists of a rejection model, feature acquisition, and a model cascade to actively augment the feature space and discover potentially unknown labels.\"}]","Exploratory Machine Learning with Unknown Unknowns | PDF",53]