[{"data":1,"prerenderedAt":-1},["ShallowReactive",2],{"doc-detail-117102-en":3,"doc-seo-117102-105":30,"detail-sidebar-cat-0-en-105":92},{"code":4,"msg":5,"data":6},0,"success",{"doc_id":7,"user_id":8,"nickname":9,"user_avatar":10,"doc_module":4,"category_id":11,"category_name":12,"doc_title":13,"doc_description":14,"doc_content":15,"file_id":16,"file_url":17,"file_type":18,"file_size":19,"view_count":20,"is_deleted":4,"is_public":20,"is_downloadable":20,"audit_status":20,"page_count":21,"language":22,"language_code":23,"site_id":24,"html_lang":23,"table_of_contents":25,"faqs":26,"seo_title":27,"seo_description":14,"update_tm":28,"read_time":29},117102,4398048949847,"Eliana","https://ap-avatar.wpscdn.com/avatar/400002536579ef2da7f?_k=1778318612642679267",8,"Research & Report","Reliability in Machine Learning - Accessible Introduction and Research Directions","Reliability in machine learning is treated as a central epistemic challenge, because models that look accurate during training can fail severely in real deployment and in changing socio-technical contexts. The work unifies research threads across statistical learning theory and philosophy of science, organizing threats during model development, deployment, and environmental adaptation. It also provides an accessible introduction to reliability-related concepts in statistics and machine learning, pointing to promising future directions.","Penultimate draft – forthcoming in Philosophy Compass.  \nReliability in Machine Learning  \nAuthors: Thomas Grote (Tübingen)  \nKonstantin Genin (Tübingen)  \nEmily Sullivan (Utrecht)  \nAbstract: Issues of reliability are claiming center-stage in the epistemology of machine learning. This paper unifies different branches in the literature and points to promising research directions, whilst also providing an accessible introduction to key concepts in statistics and machine learning  \n– as far as they are concerned with reliability.  \n.1. Introduction  \nMachine learning models often achieve impressive accuracy under training conditions, but fail in spectacular or unexpected ways when they are deployed in real-world settings. Is there some way to guarantee that predictive accuracy in training carries over to the settings in which models are actually deployed? In other words: can we be justified in relying on machine learning models on the basis of their performance in training?  \nThe underlying challenge is that there are various threats to reliability, arising at the time of (i) model development, (ii) model deployment and (iii) adapting the socio-technical environment to accommodate the model. Concerning (i), unlike traditional statistical models, there is no widely accepted mathematical theory that explains why and when state-of-the-art models such as deep neural networks generalize well. As for (ii), machine learning models are commonly used in unstable environments, or even induce changes to the environment itself, and they can be fooled by humanly imperceptible manipulations to the data. Regarding (iii), one basic issue is to align the  \nmodel output with the existing epistemic norms in a given domain, which often involves aggregating the model output with other kinds of evidence. In addition, there is another overarching problem: machine learning models are notoriously opaque, which is why it is difficult to understand the inner logic of how and why they arrive at a given prediction.  \nWhat is the relationship between the individual threats to reliability? When evaluating machine learning models, what types of assurances should we prioritize? What does the alleged (un)reliability tell us about the epistemic status of machine learning models in science and society? Can failure cases in machine learning even animate conceptual expansions or revisions of existing epistemological theories? These questions, among others, claim center-stage in the emerging debate on reliability in machine learning, spanning across statistical learning theory, mainstream epistemology, and the modeling literature in philosophy of science. This paper unifies different branches in the epistemology of machine learning and points to promising research directions, whilst also providing an accessible introduction to key concepts in statistics and machine learning – as far as they are concerned with reliability.  \n2. Statistical Learning Theory: The Received View  \nStatistical learning theory is the theory of reliability that is native to machine learning. The basics of the theory were developed by Soviet mathematicians in the 1960s; by the 1990s it was a fundamental part of the education of machine learning researchers. The spectacular successes chalked up by deep learning since the 2010s is somewhat incongruous with the theoretical predictions made by statistical learning theory. Indeed, reconciling the theoretical predictions of  \nthe canonical theory with the empirical successes of deep learningi remains one of the deepest  \ntheoretical problems in machine learning (Belkin et al., 2019) . Lack of progress in this area has caused the prestige of statistical learning theory to suffer somewhat in recent years. Nevertheless, mastering the basics of the received view is essential for understanding how machine learners approach issues of reliability.ii Statistical learning theory is a theory of an “ideal case” in which threats to reliability are well-understo","cbCaihr27fFDnUbZ","https://ap.wps.com/l/cbCaihr27fFDnUbZ","pdf",283093,1,20,"English","en",105,"# Introduction\n## Reliability threats across development, deployment, and adaptation\n# Statistical Learning Theory: The Received View\n## Supervised learning setup and ideal-case guarantees","[{\"question\":\"Why does training accuracy not guarantee real-world reliability?\",\"answer\":\"Models can perform well under training assumptions but fail in unexpected ways after deployment. Reliability threats arise from changes between training conditions and real socio-technical settings.\"},{\"question\":\"What kinds of threats to reliability does the paper distinguish?\",\"answer\":\"It groups threats according to three stages: model development, model deployment, and adapting the socio-technical environment to accommodate the model. Each stage introduces different reliability challenges.\"},{\"question\":\"How does statistical learning theory contribute to reliability analysis?\",\"answer\":\"Statistical learning theory is presented as the “received view” of reliability in machine learning, offering an ideal-case framework where threats are well-understood and can be controlled. The paper emphasizes the need to master these basics while addressing departures from classical assumptions.\"}]","Reliability in Machine Learning - Accessible Introduction and Research Directions | PDF",1785673756,50,{"code":4,"msg":31,"data":32},"ok",{"site_id":24,"language":23,"slug":33,"title":13,"keywords":34,"description":14,"schema_data":35,"social_meta":87,"head_meta":89,"extra_data":91,"updated_unix":28},"reliability-in-machine-learning-accessible-introduction-and-research-directions","",{"@graph":36,"@context":86},[37,54,69],{"@type":38,"itemListElement":39},"BreadcrumbList",[40,44,48,51],{"item":41,"name":42,"@type":43,"position":20},"https://docshare.wps.com","Home","ListItem",{"item":45,"name":46,"@type":43,"position":47},"https://docshare.wps.com/document/","Document",2,{"item":49,"name":12,"@type":43,"position":50},"https://docshare.wps.com/document/research-report/",3,{"item":52,"name":13,"@type":43,"position":53},"https://docshare.wps.com/document/reliability-in-machine-learning-accessible-introduction-and-research-directions/117102/",4,{"url":52,"name":13,"@type":55,"author":56,"headline":13,"publisher":58,"fileFormat":61,"inLanguage":23,"description":14,"dateModified":62,"datePublished":63,"encodingFormat":61,"isAccessibleForFree":64,"interactionStatistic":65},"DigitalDocument",{"name":9,"@type":57},"Person",{"url":41,"name":59,"@type":60},"DocShare","Organization","application/pdf","2026-08-05","2026-08-02",true,{"@type":66,"interactionType":67,"userInteractionCount":20},"InteractionCounter",{"@type":68},"ViewAction",{"@type":70,"mainEntity":71},"FAQPage",[72,78,82],{"name":73,"@type":74,"acceptedAnswer":75},"Why does training accuracy not guarantee real-world reliability?","Question",{"text":76,"@type":77},"Models can perform well under training assumptions but fail in unexpected ways after deployment. Reliability threats arise from changes between training conditions and real socio-technical settings.","Answer",{"name":79,"@type":74,"acceptedAnswer":80},"What kinds of threats to reliability does the paper distinguish?",{"text":81,"@type":77},"It groups threats according to three stages: model development, model deployment, and adapting the socio-technical environment to accommodate the model. Each stage introduces different reliability challenges.",{"name":83,"@type":74,"acceptedAnswer":84},"How does statistical learning theory contribute to reliability analysis?",{"text":85,"@type":77},"Statistical learning theory is presented as the “received view” of reliability in machine learning, offering an ideal-case framework where threats are well-understood and can be controlled. The paper emphasizes the need to master these basics while addressing departures from classical assumptions.","https://schema.org",{"og:url":52,"og:type":88,"og:title":13,"og:site_name":59,"og:description":14},"article",{"robots":90,"canonical":52},"index,follow",{"doc_id":7,"site_id":24},{"code":4,"msg":5,"data":93},[94,98,102,106,111,115,120,123,127,130,134],{"id":20,"doc_module":4,"doc_module_name":46,"category_name":95,"show_sort_weight":96,"slug":97},"Story & Novel",90,"story-novel",{"id":47,"doc_module":4,"doc_module_name":46,"category_name":99,"show_sort_weight":100,"slug":101},"Literature",80,"literature",{"id":53,"doc_module":4,"doc_module_name":46,"category_name":103,"show_sort_weight":104,"slug":105},"Exam",70,"exam",{"id":107,"doc_module":4,"doc_module_name":46,"category_name":108,"show_sort_weight":109,"slug":110},5,"Comic",60,"comic",{"id":112,"doc_module":4,"doc_module_name":46,"category_name":113,"show_sort_weight":29,"slug":114},6,"Technology","technology",{"id":116,"doc_module":4,"doc_module_name":46,"category_name":117,"show_sort_weight":118,"slug":119},7,"Healthcare",40,"healthcare",{"id":11,"doc_module":4,"doc_module_name":46,"category_name":12,"show_sort_weight":121,"slug":122},30,"research-report",{"id":124,"doc_module":4,"doc_module_name":46,"category_name":125,"show_sort_weight":21,"slug":126},9,"Religion & Spirituality","religion-spirituality",{"id":21,"doc_module":4,"doc_module_name":46,"category_name":128,"show_sort_weight":21,"slug":129},"World Cup","world-cup",{"id":131,"doc_module":4,"doc_module_name":46,"category_name":132,"show_sort_weight":131,"slug":133},10,"Lifestyle","lifestyle",{"id":135,"doc_module":4,"doc_module_name":46,"category_name":136,"show_sort_weight":107,"slug":137},19,"General","general"]