Zhihao Hu, Yiran Xu, Mengnan Du, Jindong Gu, Xinmei Tian, Fengxiang He
{"title":"Boosting Fair Classifier Generalization through Adaptive Priority Reweighing","authors":"Zhihao Hu, Yiran Xu, Mengnan Du, Jindong Gu, Xinmei Tian, Fengxiang He","doi":"10.1145/3665895","DOIUrl":null,"url":null,"abstract":"<p>With the increasing penetration of machine learning applications in critical decision-making areas, calls for algorithmic fairness are more prominent. Although there have been various modalities to improve algorithmic fairness through learning with fairness constraints, their performance does not generalize well in the test set. A performance-promising fair algorithm with better generalizability is needed. This paper proposes a novel adaptive reweighing method to eliminate the impact of the distribution shifts between training and test data on model generalizability. Most previous reweighing methods propose to assign a unified weight for each (sub)group. Rather, our method granularly models the distance from the sample predictions to the decision boundary. Our adaptive reweighing method prioritizes samples closer to the decision boundary and assigns a higher weight to improve the generalizability of fair classifiers. Extensive experiments are performed to validate the generalizability of our adaptive priority reweighing method for accuracy and fairness measures (i.e., equal opportunity, equalized odds, and demographic parity) in tabular benchmarks. We also highlight the performance of our method in improving the fairness of language and vision models. The code is available at https://github.com/che2198/APW.</p>","PeriodicalId":49249,"journal":{"name":"ACM Transactions on Knowledge Discovery from Data","volume":"21 1","pages":""},"PeriodicalIF":4.0000,"publicationDate":"2024-05-23","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":"0","resultStr":null,"platform":"Semanticscholar","paperid":null,"PeriodicalName":"ACM Transactions on Knowledge Discovery from Data","FirstCategoryId":"94","ListUrlMain":"https://doi.org/10.1145/3665895","RegionNum":3,"RegionCategory":"计算机科学","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":null,"EPubDate":"","PubModel":"","JCR":"Q1","JCRName":"COMPUTER SCIENCE, INFORMATION SYSTEMS","Score":null,"Total":0}
引用次数: 0
Abstract
With the increasing penetration of machine learning applications in critical decision-making areas, calls for algorithmic fairness are more prominent. Although there have been various modalities to improve algorithmic fairness through learning with fairness constraints, their performance does not generalize well in the test set. A performance-promising fair algorithm with better generalizability is needed. This paper proposes a novel adaptive reweighing method to eliminate the impact of the distribution shifts between training and test data on model generalizability. Most previous reweighing methods propose to assign a unified weight for each (sub)group. Rather, our method granularly models the distance from the sample predictions to the decision boundary. Our adaptive reweighing method prioritizes samples closer to the decision boundary and assigns a higher weight to improve the generalizability of fair classifiers. Extensive experiments are performed to validate the generalizability of our adaptive priority reweighing method for accuracy and fairness measures (i.e., equal opportunity, equalized odds, and demographic parity) in tabular benchmarks. We also highlight the performance of our method in improving the fairness of language and vision models. The code is available at https://github.com/che2198/APW.
期刊介绍:
TKDD welcomes papers on a full range of research in the knowledge discovery and analysis of diverse forms of data. Such subjects include, but are not limited to: scalable and effective algorithms for data mining and big data analysis, mining brain networks, mining data streams, mining multi-media data, mining high-dimensional data, mining text, Web, and semi-structured data, mining spatial and temporal data, data mining for community generation, social network analysis, and graph structured data, security and privacy issues in data mining, visual, interactive and online data mining, pre-processing and post-processing for data mining, robust and scalable statistical methods, data mining languages, foundations of data mining, KDD framework and process, and novel applications and infrastructures exploiting data mining technology including massively parallel processing and cloud computing platforms. TKDD encourages papers that explore the above subjects in the context of large distributed networks of computers, parallel or multiprocessing computers, or new data devices. TKDD also encourages papers that describe emerging data mining applications that cannot be satisfied by the current data mining technology.