[{"data":1,"prerenderedAt":-1},["ShallowReactive",2],{"doc-detail-122169-en":3,"doc-seo-122169-105":30,"detail-sidebar-cat-0-en-105":84},{"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},122169,7971461740886,"Theodore","https://ap-avatar.wpscdn.com/davatar_3d24733baf745e90a7e4bdd5f77d97b2",8,"Research & Report","Fair Representation Learning with Unreliable Labels","Fair representation learning faces label bias in which practitioners’ prejudice systematically flips labels between classes, distorting how predictiveness is measured. Existing approaches often aim to remove dependence between sensitive attributes and inputs, but fail to ensure learned representations stay informative when training labels are unreliable. The paper shows representations can become random or degenerate under biased contamination, then proposes a disentangling latent model that enforces independence from sensitive factors using a mutual-information penalty while preserving task-relevant signal. Experiments on synthetic and real datasets verify improved performance over prior methods.","Fair Representation Learning with Unreliable Labels  \nYixuan Zhang†, Feng Zhou‡∗ , Zhidong Li†, Yang Wang†, Fang Chen††Data Science Institute, University of Technology Sydney, Australia ‡Center for Applied Statistics and School of Statistics, Renmin University of China, China  \n{yixuan.zhang, [zhidong.li](zhidong.li) , [yang.wang](yang.wang) , [fang.chen](fang.chen}@uts.edu.au)[}](fang.chen}@uts.edu.au)[@uts.edu.au](fang.chen}@uts.edu.au), [feng.zhou@ruc.edu.cn](feng.zhou@ruc.edu.cn)  \nAbstract  \nIn learning with fairness, for every instance, its label can be systematically flipped to another class due to the practitioner’s prejudice, namely, label bias. The existing well-studied fair representation learning methods focus on removing the dependency between the sensitive factors and the input data, but do not address how the representations retain useful information when the labels are unreliable. In fact, we find that the learned representations become random or degenerated when the instance is contaminated by label bias. To alleviate this issue, we investigate the problem of learning fair representations that are independent of the sensitive factors while retaining the task-relevant information given only access to unreliable labels. Our model disentangles the dependency between fair representations and sensitive factors in the latent space. To remove the reliance between the labelsand sensitive factors, we incorporate an additional penalty based on mutual information. The learned purged fair representations can then be used in any downstream processing. We demonstrate the superiority of our method over previous works through multiple experiments on both synthetic and real-world datasets.  \n1 Introduction  \nThe recent success of deploying machine learning algorithms in different high-stake application areas has increased the concerns for ethics. Due to human prejudice intervening in the labeling process, the training data collected always contains discrimination towards certain de-  \nProceedings of the 26th International Conference on Artificial Intelligence and Statistics (AISTATS) 2023, Valencia, Spain. PMLR: Volume 206 . Copyright 2023 by the author(s) .  \n*Corresponding author.  \n0.6  \n0.4  \n0.2  \n0.0  \n−0.2  \n−0.4  \n−0.6  \nz vPCsAyon the latent representationz v.s.   \n−0.5 0.0 0.5 −0.5 0.0 0.5  \nFigure 1: Principal component analysis (PCA) on the learned representations. We performed a PCA analysis between the learned representation and label for a binary classification problem on the synthetic dataset. Red points represent the positive class, while blue points represent the negative class. We compare the ideal labels (left) with unreliable labels (right): the learned representation has a strong correlation with the ideal label (obviously divided into two clusters) but a weak correlation with the unreliable label (two clusters mixed together) .  \nmographic groups (Lin et al., 2020 ; Bertrand and Mullainathan, 2004 ; Michelle, 2012) . When decisions are made algorithmically with such unreliable labels, it affects both accuracy and fairness negatively, and further brings harm to both society and individuals (Khandani et al., 2010 ; Kim et al., 2015 ; Brennan et al., 2009) . Therefore, as one critical ethical aspect, fairness-aware learning has recently experienced a surge of advances.  \nExisting works on fairness extensively studied discrimination removal strategies in different training stages, i.e., pre-processing (Louizos et al., 2015a ; Zemel et al., 2013 ; Calmon et al., 2017 ; Lum and Johndrow, 2016), inprocessing (Bilal Zafar et al., 2015, 2016 ; Calders et al., 2009 ; Agarwal et al., 2018 ; Kamishima et al., 2012) and post-processing (Hardt et al., 2016) . Among all these methods, fair representation learning (Louizos et al., 2015b ; Creager et al., 2019 ; Zemel et al., 2013 ; Calmon et al., 2017) as a pre-processing method has gained significant attention because it is compatible with any learning algo-  \nrithms f","cbCaih9JrTqcCkKv","https://ap.wps.com/l/cbCaih9JrTqcCkKv","pdf",2997668,1,13,"English","en",105,"# Introduction\n## Motivation: fairness and label bias\n## Background: fairness-aware learning and fair representation learning\n## Problem formulation and research question\n## Proposed approach using VAE and mutual information","[{\"question\":\"How is performance evaluated?\",\"answer\":\"The paper demonstrates superiority through multiple experiments on both synthetic and real-world datasets.\"}]","Fair Representation Learning with Unreliable Labels | PDF",1785809166,33,{"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":79,"head_meta":81,"extra_data":83,"updated_unix":28},"fair-representation-learning-with-unreliable-labels","",{"@graph":36,"@context":78},[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/fair-representation-learning-with-unreliable-labels/122169/",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","2026-08-04",true,{"@type":66,"interactionType":67,"userInteractionCount":20},"InteractionCounter",{"@type":68},"ViewAction",{"@type":70,"mainEntity":71},"FAQPage",[72],{"name":73,"@type":74,"acceptedAnswer":75},"How is performance evaluated?","Question",{"text":76,"@type":77},"The paper demonstrates superiority through multiple experiments on both synthetic and real-world datasets.","Answer","https://schema.org",{"og:url":52,"og:type":80,"og:title":13,"og:site_name":59,"og:description":14},"article",{"robots":82,"canonical":52},"index,follow",{"doc_id":7,"site_id":24},{"code":4,"msg":5,"data":85},[86,90,94,98,103,108,113,116,121,124,128],{"id":20,"doc_module":4,"doc_module_name":46,"category_name":87,"show_sort_weight":88,"slug":89},"Story & Novel",90,"story-novel",{"id":47,"doc_module":4,"doc_module_name":46,"category_name":91,"show_sort_weight":92,"slug":93},"Literature",80,"literature",{"id":53,"doc_module":4,"doc_module_name":46,"category_name":95,"show_sort_weight":96,"slug":97},"Exam",70,"exam",{"id":99,"doc_module":4,"doc_module_name":46,"category_name":100,"show_sort_weight":101,"slug":102},5,"Comic",60,"comic",{"id":104,"doc_module":4,"doc_module_name":46,"category_name":105,"show_sort_weight":106,"slug":107},6,"Technology",50,"technology",{"id":109,"doc_module":4,"doc_module_name":46,"category_name":110,"show_sort_weight":111,"slug":112},7,"Healthcare",40,"healthcare",{"id":11,"doc_module":4,"doc_module_name":46,"category_name":12,"show_sort_weight":114,"slug":115},30,"research-report",{"id":117,"doc_module":4,"doc_module_name":46,"category_name":118,"show_sort_weight":119,"slug":120},9,"Religion & Spirituality",20,"religion-spirituality",{"id":119,"doc_module":4,"doc_module_name":46,"category_name":122,"show_sort_weight":119,"slug":123},"World Cup","world-cup",{"id":125,"doc_module":4,"doc_module_name":46,"category_name":126,"show_sort_weight":125,"slug":127},10,"Lifestyle","lifestyle",{"id":129,"doc_module":4,"doc_module_name":46,"category_name":130,"show_sort_weight":99,"slug":131},19,"General","general"]