[{"data":1,"prerenderedAt":-1},["ShallowReactive",2],{"doc-seo-200529-105":3,"detail-sidebar-cat-1-en-105":80,"doc-detail-200529-en":126},{"code":4,"msg":5,"data":6},0,"ok",{"site_id":7,"language":8,"slug":9,"title":10,"keywords":11,"description":12,"schema_data":13,"social_meta":73,"head_meta":75,"extra_data":77,"updated_unix":79},105,"en","explainability-fairness-and-evaluation-in-machine-learning-from-theory-to-policy-and-back","Explainability, fairness, and evaluation in machine learning: From theory to policy and back","","Machine learning models trained on historical data can cause harmful, hard-to-detect decisions about individuals, driven by issues such as biased training data, implementation bugs, and generalization failures. The thesis evaluates “fair machine learning” and explainable AI approaches through quantitative and qualitative analysis. It examines explainability—especially feature importance—using game-theoretic principles, then frames key harms and failure modes as public policy concerns. It further derives data-dependent group-specific generalization bounds to explain performance disparities across heterogeneous groups.",{"@graph":14,"@context":72},[15,34,55],{"@type":16,"itemListElement":17},"BreadcrumbList",[18,23,27,31],{"item":19,"name":20,"@type":21,"position":22},"https://docshare.wps.com","Home","ListItem",1,{"item":24,"name":25,"@type":21,"position":26},"https://docshare.wps.com/template/","Template",2,{"item":28,"name":29,"@type":21,"position":30},"https://docshare.wps.com/template/presentations/","Presentations",3,{"item":32,"name":10,"@type":21,"position":33},"https://docshare.wps.com/template/explainability-fairness-and-evaluation-in-machine-learning-from-theory-to-policy-and-back/200529/",4,{"url":32,"name":10,"@type":35,"image":36,"author":41,"headline":10,"publisher":44,"fileFormat":47,"inLanguage":8,"description":12,"dateModified":48,"datePublished":49,"encodingFormat":47,"isAccessibleForFree":50,"interactionStatistic":51},"DigitalDocument",{"url":37,"@type":38,"width":39,"height":40},"https://docshare.wps.com/thumbnails/explainability-fairness-and-evaluation-in-machine-learning-from-theory-to-policy-and-back/200529.png","ImageObject",442,249,{"name":42,"@type":43},"Miles","Person",{"url":19,"name":45,"@type":46},"DocShare","Organization","application/pdf","2026-10-11","2026-09-04",true,{"@type":52,"interactionType":53,"userInteractionCount":33},"InteractionCounter",{"@type":54},"ViewAction",{"@type":56,"mainEntity":57},"FAQPage",[58,64,68],{"name":59,"@type":60,"acceptedAnswer":61},"Why do machine learning models pose risks when used to make decisions about people?","Question",{"text":62,"@type":63},"Models can produce harmful decisions while remaining unnoticed due to complexity and opacity. Risks arise from biased training data, bugs, and failure modes including underfitting or overfitting.","Answer",{"name":65,"@type":60,"acceptedAnswer":66},"What does the thesis do to assess explainability methods?",{"text":67,"@type":63},"It focuses on feature importance techniques and evaluates specific algorithms using game-theoretic principles, combining quantitative and qualitative assessments. It also interrogates the philosophy behind the methodology.",{"name":69,"@type":60,"acceptedAnswer":70},"How are machine learning harms treated in the thesis?",{"text":71,"@type":63},"The thesis argues that certain harms and failure modes must be treated as public policy issues. It also offers suggestions for how to address them.","https://schema.org",{"og:url":32,"og:type":74,"og:title":10,"og:site_name":45,"og:description":12},"article",{"robots":76,"canonical":32},"index,follow",{"doc_id":78,"site_id":7},200529,1788509229,{"code":4,"msg":81,"data":82},"success",[83,87,92,97,102,107,112,117,122],{"id":84,"doc_module":22,"doc_module_name":25,"category_name":29,"show_sort_weight":85,"slug":86},11,90,"presentations",{"id":88,"doc_module":22,"doc_module_name":25,"category_name":89,"show_sort_weight":90,"slug":91},12,"Resumes",80,"resumes",{"id":93,"doc_module":22,"doc_module_name":25,"category_name":94,"show_sort_weight":95,"slug":96},14,"Invoices",70,"invoices",{"id":98,"doc_module":22,"doc_module_name":25,"category_name":99,"show_sort_weight":100,"slug":101},15,"Posters",60,"posters",{"id":103,"doc_module":22,"doc_module_name":25,"category_name":104,"show_sort_weight":105,"slug":106},16,"Social Media",50,"social-media",{"id":108,"doc_module":22,"doc_module_name":25,"category_name":109,"show_sort_weight":110,"slug":111},17,"Forms",40,"forms",{"id":113,"doc_module":22,"doc_module_name":25,"category_name":114,"show_sort_weight":115,"slug":116},18,"Letters",30,"letters",{"id":118,"doc_module":22,"doc_module_name":25,"category_name":119,"show_sort_weight":120,"slug":121},21,"Paper Templates",5,"papers-templates",{"id":123,"doc_module":22,"doc_module_name":25,"category_name":124,"show_sort_weight":4,"slug":125},158,"General","general-158",{"code":4,"msg":81,"data":127},{"doc_id":78,"user_id":128,"nickname":42,"user_avatar":129,"doc_module":22,"category_id":84,"category_name":29,"doc_title":10,"doc_description":12,"doc_content":130,"file_id":131,"file_url":132,"file_type":133,"file_size":134,"view_count":135,"is_deleted":4,"is_public":22,"is_downloadable":22,"audit_status":22,"page_count":136,"language":137,"language_code":8,"site_id":7,"html_lang":8,"table_of_contents":138,"faqs":139,"seo_title":140,"seo_description":12,"update_tm":79,"read_time":141},13056703019404,"https://ap-avatar.wpscdn.com/davatar_29158cc5080c5b710cf443261637dec0","Abstract of “Explainability, fairness, and evaluation in machine learning: From theory to policy and back” by Indra Elizabeth Kumar, Ph.D., Brown University, May 2024 .  \nMachine learning models, or algorithms trained on historical data, can be harmful when they are used to make decisions about people. From underfitting to overfitting, from bugs in code to biased training data, machine learning engineers and other stakeholders alike are concerned with the capacity of these algorithms to make problematic decisions while going unnoticed by their human supervisors, escaping scrutiny due to their increasing complexity and incomprehensibility. In recent years, this has led to a flurry of technical research on methods to anticipate, detect, or prevent undesirable behavior, under the umbrella terms of “fair machine learning” and “explainable artificial intelligence.” But these efforts have not gone without criticism, especially from social science disciplines quick to point out problems with a techno-solutionist approach. In this thesis, we concern ourselves with trying to articulate what exactly we can learn from quantitative methods proposed to address this problem and what we should do with that information. In the first Part, which centers on a family of “explainability” techniques known as feature importance, we quantitatively and qualitatively evaluate specific algorithms based on game theoretic principles, and interrogate the philosophy behind the overall methodology. The next Part articulates and draws attention to specific machine learning harms and failure modes that must be treated as public policy issues and offer suggestions on how to do so. For the third Part, we derive novel, data-dependent, group-specific generalization bounds on the performance of machine learning models trained on heterogeneous data, which offers a framework to understand the data and modeling conditions which cause performance disparities across those groups.  \nExplainability, fairness, and evaluation in machine learning: From theory to policy and back  \nby  \nIndra Elizabeth Kumar  \nB. A., Scripps College, 2016  \nM. S., University of Massachusetts, 2019  \nA dissertation submitted in partial fulfillment of the  \nrequirements for the Degree of Doctor of Philosophy in the Department of Computer Science at Brown University  \nProvidence, Rhode Island  \nMay 2024  \n© Copyright 2024 by Indra Elizabeth Kumar  \nThis dissertation by Indra Elizabeth Kumar is accepted in its present form by the Department of Computer Science as satisfying the dissertation requirement for the degree of Doctor of Philosophy.  \nDate      \nSuresh Venkatasubramanian, Director  \nRecommended to the Graduate Council  \nDate      \nTimothy Edgar, Reader  \nDate      \nStephen Bach, Reader  \nApproved by the Graduate Council  \nDate      \nThomas A. Lewis  \nDean of the Graduate School  \nAcknowledgements  \nI am grateful to the many wonderful people and institutions that supported this work, who I’ve done my best to enumerate here.  \nFirst, I want to thank Suresh for being everything I could have hoped for in an advisor. Your guidance and support have been my greatest assets throughout my research career thus far. Thankyou for trusting me to make my own decisions, pushing me to do my best work, and prioritizing my wellbeing. I would also like to thank the rest of my committee for their thoughtful feedback on this work, as well as the many other members of the faculty and staff at the Utah School of Computing, Brown Data Science Institute, and Brown Computer Science Department who have taught me things, provided support, and just generally helped me out over the years. I feel lucky to have been affiliated with both institutions during my time as a graduate student.  \nThis dissertation would not have been possible without my excellent coauthors on the work it contains; it has been an honor to collaborate with and learn from all of you. Similarly, to my fellow graduate students at the U who started ","cbCait93hMwCj3zQ","https://ap.wps.com/l/cbCait93hMwCj3zQ","pdf",2310902,10,165,"English","# Introduction\n## Understanding feature importance and its limitations\n### On the limitations of Shapley value explanations\n### Quantifying the limitations of Shapley value explanations\n### Using the lens of feminist epistemology to critique feature importance\n## Public policy problems in machine learning\n### Applying the Equal Credit Opportunity Act to machine learning models","[{\"question\":\"Why do machine learning models pose risks when used to make decisions about people?\",\"answer\":\"Models can produce harmful decisions while remaining unnoticed due to complexity and opacity. Risks arise from biased training data, bugs, and failure modes including underfitting or overfitting.\"},{\"question\":\"What does the thesis do to assess explainability methods?\",\"answer\":\"It focuses on feature importance techniques and evaluates specific algorithms using game-theoretic principles, combining quantitative and qualitative assessments. It also interrogates the philosophy behind the methodology.\"},{\"question\":\"How are machine learning harms treated in the thesis?\",\"answer\":\"The thesis argues that certain harms and failure modes must be treated as public policy issues. It also offers suggestions for how to address them.\"}]","Explainability, fairness, and evaluation in machine learning: From theory to policy and back | PDF",58]