[{"data":1,"prerenderedAt":-1},["ShallowReactive",2],{"doc-detail-117954-en":3,"doc-seo-117954-105":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},117954,3848291630094,"Emma Wilson","https://eur-avatar.wpscdn.com/davatar_085a072bc5b1113ac321206ff7593b45",8,"Research & Report","Machine Learning and Knowledge - Why Robustness Matters","Trust in machine learning depends on confidence in outputs, usually expressed through reliability as the rate of correct predictions. Reliability alone misses robustness concerns, including reliance on irrelevant features and performance changes under contextual shifts. The paper argues that the epistemic dimension of trust is best captured by knowledge: users are trustworthy only when they are positioned to know outputs are correct. Knowledge requires appropriately grounded beliefs that remain robust to error, which yields criteria based on counterfactual success and correct feature dependence.","arXiv :2310 . 19819v1 [ cs .LG] 23 Oct 2023  \nMachine Learning and Knowledge: Why Robustness  \nMatters  \nJonathan Vandenburgh  \nAbstract  \nTrusting machine learning algorithms requires having confidence in their outputs.  \nConfidence is typically interpreted in terms of model reliability, where a model is reliable if it produces a high proportion of correct outputs. However, model reliability does not address concerns about the robustness of machine learning models, such as models relying on the wrong features or variations in performance based on context. I argue that the epistemic dimension of trust can instead be understood through the concept of knowledge, where the trustworthiness of an algorithm depends on whether its users are in the position to know that its outputs are correct. Knowledge requires beliefs to be formed for the right reasons and to be robust to error, so machine learning algorithms can only provide knowledge if they work well across counterfactual scenarios and if they make decisions based on the right features. This, I argue, can explain why we should care about model properties like interpretability, causal shortcut independence, and distribution shift robustness even if such properties are not required for model reliability.  \nSuppose, following DeGrave et al. (2021), that a hospital trains a machine learning model to detect whether a patient has Covid-19 based on chest x-ray images. The hospital trains this model with publicly available data, including Covid-negative x-rays from an NIH dataset and Covid-positive x-rays from a dataset found on Github, and achieves excellent performance when testing the model. However, unbeknownst to the hospital, part of the reason why the model appears successful is that it has learned to discern differences other than Covid status between the images in the NIH dataset and the Github dataset. For example, x-rays in the NIH dataset of Covid-positive patients tend to have more space above a patient’s shoulders, and by learning this correlation, the machine learning model is able to improve performance in identifying Covid in the training environment.  \nEven though the evidence available to the hospital suggests that the model works incredibly well, and even though it may work reasonably well in practice, it seems inappropriate for the hospital to rely on the algorithm for Covid diagnosis. Some guidance as to why this model is untrustworthy can be found in work on the robustness of machine learning models. First, it does not generalize well: while it performs well in the training context, it performs less well in a new environment where the data looks different than it does in training. Second, it fails an interpretability test: when using an algorithm to interpret how the model predicts Covid status, the model appears to be focused on irrelevant parts of the x-ray image, such as the space above a patient’s shoulders. Third, it relies on a shortcut or confounder, making a decision based on a feature which is spuriously correlated with Covid status in the training data rather than based on features that are causally relevant to having Covid.  \nWhile these seem like compelling reasons not to trust the Covid detection algorithm described above, it is unclear how well these reasons generalize or how they fit into abroader conception of trustworthy machine learning. For example, some authors argue that interpretability is not necessary for trusting algorithms, since it is either very difficult or impossible to meaningfully interpret the inner workings of a “black box” machine learning algorithm (Dur´an and Formanek, 2018; Krishnan, 2020) . Further, looking toward the dominant epistemic theory associated with trustworthiness, computational reliabilism (Dur´an and Formanek, 2018; Dur´an and Jongsma, 2021), the above factors are either irrelevant for trustworthiness or relevant only insofar as they are indicators of reliable performance (Mishra, 2021) .  \nAnother avenue for add","cbCaivsXvf7J3cxO","https://ap.wps.com/l/cbCaivsXvf7J3cxO","pdf",327853,1,16,"English","en",105,"# Introduction\n## Reliability vs robustness\n# Knowledge-based account of trust\n## Robustness across counterfactual scenarios\n## Correct feature dependence\n# Case study: Covid-19 X-ray detection\n## Dataset mismatch and shortcut learning\n## Failure modes: generalization, interpretability, confounders\n# Implications and evaluation of trustworthy ML\n## Interpretability and causal shortcut independence\n## Distribution shift robustness\n# Ethical considerations and limits\n# Conclusion and paper roadmap","[{\"question\":\"Why is model reliability not enough for trustworthy machine learning?\",\"answer\":\"Reliability focuses on how often outputs are correct, but it does not address robustness issues such as learning the wrong features or changing behavior across contexts and distributions.\"},{\"question\":\"How does the paper define the epistemic dimension of trustworthiness?\",\"answer\":\"Trustworthiness is tied to knowledge: users must be in a position to know that the algorithm’s outputs are correct, where beliefs are formed for the right reasons and remain robust to error.\"},{\"question\":\"What does the Covid-19 X-ray example show about robustness failures?\",\"answer\":\"The model performs well in testing but may rely on dataset-specific differences rather than Covid status, demonstrating failures in generalization, interpretability, and causal feature dependence.\"}]","Machine Learning and Knowledge - Why Robustness Matters | PDF",1785680519,40,{"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},"machine-learning-and-knowledge-why-robustness-matters","",{"@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/machine-learning-and-knowledge-why-robustness-matters/117954/",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 is model reliability not enough for trustworthy machine learning?","Question",{"text":76,"@type":77},"Reliability focuses on how often outputs are correct, but it does not address robustness issues such as learning the wrong features or changing behavior across contexts and distributions.","Answer",{"name":79,"@type":74,"acceptedAnswer":80},"How does the paper define the epistemic dimension of trustworthiness?",{"text":81,"@type":77},"Trustworthiness is tied to knowledge: users must be in a position to know that the algorithm’s outputs are correct, where beliefs are formed for the right reasons and remain robust to error.",{"name":83,"@type":74,"acceptedAnswer":84},"What does the Covid-19 X-ray example show about robustness failures?",{"text":85,"@type":77},"The model performs well in testing but may rely on dataset-specific differences rather than Covid status, demonstrating failures in generalization, interpretability, and causal feature dependence.","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},[94,98,102,106,111,116,120,123,128,131,135],{"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":114,"slug":115},6,"Technology",50,"technology",{"id":117,"doc_module":4,"doc_module_name":46,"category_name":118,"show_sort_weight":29,"slug":119},7,"Healthcare","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":126,"slug":127},9,"Religion & Spirituality",20,"religion-spirituality",{"id":126,"doc_module":4,"doc_module_name":46,"category_name":129,"show_sort_weight":126,"slug":130},"World Cup","world-cup",{"id":132,"doc_module":4,"doc_module_name":46,"category_name":133,"show_sort_weight":132,"slug":134},10,"Lifestyle","lifestyle",{"id":136,"doc_module":4,"doc_module_name":46,"category_name":137,"show_sort_weight":107,"slug":138},19,"General","general"]