Assessing Classifier Fairness with Collider Bias

被引:4
作者
Xu, Zhenlong [1 ]
Xu, Ziqi [1 ]
Liu, Jixue [1 ]
Cheng, Debo [1 ]
Li, Jiuyong [1 ]
Liu, Lin [1 ]
Wang, Ke [2 ]
机构
[1] Univ South Australia, Adelaide, SA, Australia
[2] Simon Fraser Univ, Burnaby, BC, Canada
来源
ADVANCES IN KNOWLEDGE DISCOVERY AND DATA MINING, PAKDD 2022, PT II | 2022年 / 13281卷
基金
加拿大自然科学与工程研究理事会; 澳大利亚研究理事会;
关键词
Fairness; Collider bias; Causal inference;
D O I
10.1007/978-3-031-05936-0_21
中图分类号
TP18 [人工智能理论];
学科分类号
081104 ; 0812 ; 0835 ; 1405 ;
摘要
The increasing application of machine learning techniques in everyday decision-making processes has brought concerns about the fairness of algorithmic decision-making. This paper concerns the problem of collider bias which produces spurious associations in fairness assessment and develops theorems to guide fairness assessment avoiding the collider bias. We consider a real-world application of auditing a trained classifier by an audit agency. We propose an unbiased assessment algorithm by utilising the developed theorems to reduce collider biases in the assessment. Experiments and simulations show the proposed algorithm reduces collider biases significantly in the assessment and is promising in auditing trained classifiers.
引用
收藏
页码:262 / 276
页数:15
相关论文
共 25 条
[1]  
[Anonymous], 2011, P 17 ACM SIGKDD INT, DOI DOI 10.1145/2020408.2020488
[2]  
Asuncion A., 2007, Uci machine learning repository
[3]  
Bendick M., 2007, Horizons strategiques, V5, P17, DOI [DOI 10.3917/HORI.005.0017, 10.3917/hori.005.0017]
[4]  
Caton S, 2020, Arxiv, DOI [arXiv:2010.04053, 10.48550/arXiv.2010.04053]
[5]  
Chiappa S, 2019, AAAI CONF ARTIF INTE, P7801
[6]   Illustrating bias due to conditioning on a collider [J].
Cole, Stephen R. ;
Platt, Robert W. ;
Schisterman, Enrique F. ;
Chu, Haitao ;
Westreich, Daniel ;
Richardson, David ;
Poole, Charles .
INTERNATIONAL JOURNAL OF EPIDEMIOLOGY, 2010, 39 (02) :417-420
[7]  
Corbett-Davies S, 2018, Arxiv, DOI arXiv:1808.00023
[8]  
Dwork C., 2012, P 3 INN THEOR COMP S, P214, DOI DOI 10.1145/2090236.2090255
[9]   Certifying and Removing Disparate Impact [J].
Feldman, Michael ;
Friedler, Sorelle A. ;
Moeller, John ;
Scheidegger, Carlos ;
Venkatasubramanian, Suresh .
KDD'15: PROCEEDINGS OF THE 21ST ACM SIGKDD INTERNATIONAL CONFERENCE ON KNOWLEDGE DISCOVERY AND DATA MINING, 2015, :259-268
[10]  
Hardt M, 2016, ADV NEUR IN, V29