融合对抗与多目标的公平机器学习算法及可视化平台研究

机器学习技术的蓬勃发展对人类社会产生了重要的影响，各类机器学习算法在辅助决策系统的应用极大方便了人们的生活。然而，一些机器学习算法存在对部分人群的歧视性或偏向性，这可能会引起社会的不公，从而引发一系列社会问题。因此，越来越多的机器学习领域的学者在近几年开始聚焦于机器学习公平性的研究。一方面，研究者们提出了大量的公平性评价指标来量化不同场景中的公平性；另一方面，他们还研究了一系列作用于机器学习不同阶段的方法以缓解模型的偏见。此外，还有的研究者致力于整合和集成现有工作，开发公平性平台或算法库以供其他机器学习研究者或人工智能伦理工程师使用。本文基于前人的相关研究，针对公平机器学习算法和可视化平台做出了以下两方面工作。
  （1）在面向公平性的机器学习中，模型公平性的提升往往会导致准确率的降低，对两者的权衡是研究中的一大挑战。有的研究者使用了多目标演化学习算法来同时优化模型的公平性和准确率，本文基于该框架，将对抗网络作为公平性指标的代理，在反向传播的过程中将代表“公平”方向的梯度信息传递给预测器，提高对公平性的优化效率。此外，我们还利用演化算法种群中个体在适应度上的差异性设计了一种自适应策略，可以动态平衡反向传播中优化准确性和优化公平性的梯度项，以增大搜索空间。通过多个在公平性研究中被广泛使用的数据集上的实验表明，本文的算法相比于前人的算法在整体表现上有所提升，能够找到更多样、更公平的机器学习模型集。
  （2）在现有的公平性平台中，它们中的大部分都没有提供用户界面来可视化数据集和模型的分析结果，或支持用户上传自己的数据集和模型进行分析。此外，还没有平台集成了多目标算法以可视化地优化模型的准确性和多个公平性指标。为了填补上述空缺，我们开发了FairerML：一个可扩展的，用于分析、可视化和缓解机器学习中偏见的机器学习公平性平台。该平台主要实现了以下功能：对数据集和模型进行公平性分析，以多目标演化学习训练一组能满足不同公平性和准确性权衡的帕累托模型集。平台针对公平性分析和模型训练提供了多种的可视化手段，提升了平台的可用性。此外，因为集成的多目标算法能够支持任意的公平性指标，故FairerML可以集成新的指标以供分析或作为优化目标。

[1] LIEM C, LANGER M, DEMETRIOU A, et al. Psychology meets machine learning: Interdisciplinary perspectives on algorithmic job candidate screening[M]//Explainable and interpretable models in computer vision and machine learning. Springer, 2018: 197-253.
[2] FUJIYOSHI H, HIRAKAWA T, YAMASHITA T. Deep learning-based image recognition for autonomous driving[J]. IATSS Research, 2019, 43(4): 244-252.
[3] WU Y, SCHUSTER M, CHEN Z, et al. Google’s neural machine translation system: Bridging the gap between human and machine translation[J]. arXiv preprint arXiv:1609.08144, 2016.
[4] MUKERJEE A, BISWAS R, DEB K, et al. Multi–objective evolutionary algorithms for the risk–return trade–off in bank loan management[J]. International Transactions in Operational Research, 2002, 9(5): 583-597.
[5] LI L, LASSITER T, OH J, et al. Algorithmic hiring in practice: Recruiter and HR professional’s perspectives on AI use in hiring[C]//AIES ’21: Proceedings of the 2021 AAAI/ACM Conference on AI, Ethics, and Society. Association for Computing Machinery, 2021: 166–176.
[6] BRENNAN T, DIETERICH W, EHRET B. Evaluating the predictive validity of the COMPAS risk and needs assessment system[J]. Criminal Justice and Behavior, 2009, 36(1): 21-40.
[7] MEHRABI N, MORSTATTER F, SAXENA N, et al. A survey on bias and fairness in machine learning[J]. ACM Computing Surveys (CSUR), 2021, 54(6): 1-35.
[8] CORBETT-DAVIES S, PIERSON E, FELLER A, et al. Algorithmic decision making and the cost of fairness[C]//Proceedings of the 23rd ACM SIGKDD International Conference on Knowledge Discovery and Data Mining. 2017: 797-806.
[9] HUANG C W, ZHANG Z Q, MAO B F, et al. An overview of artificial intelligence ethics[J]. IEEE Transactions on Artificial Intelligence, 2022, Early Access: 1-21.
[10] 刘文炎; 沈楚云; 王祥丰; 金博; 卢兴见; 王晓玲; 查宏远; 何积丰;. 可信机器学习的公平性综述[J]. 软件学报, 2021, 32(5): 1404-1426.
[11] CATON S, HAAS C. Fairness in machine learning: A Survey[J]. arXiv preprint- arXiv:2010.04053, 2020.
[12] PESSACH D, SHMUELI E. A review on fairness in machine learning[J]. ACM Computing Surveys (CSUR), 2022, 55(3): 1-44.
[13] FELDMAN M, FRIEDLER S A, MOELLER J, et al. Certifying and removing disparate impact [C]//Proceedings of the 21th ACM SIGKDD International Conference on Knowledge Discovery and Data Mining. 2015: 259-268.
[14] KAMIRAN F, CALDERS T. Data preprocessing techniques for classification without discrimination[J]. Knowledge and Information Systems, 2012, 33(1): 1-33.
[15] KAMISHIMA T, AKAHO S, ASOH H, et al. Fairness-aware classifier with prejudice remover regularizer[C]//Joint European Conference on Machine Learning and Knowledge Discovery in Databases. Springer, 2012: 35-50.
[16] GOH G, COTTER A, GUPTA M, et al. Satisfying real-world goals with dataset constraints[J]. Advances in Neural Information Processing Systems, 2016, 29.
[17] ZHANG B H, LEMOINE B, MITCHELL M. Mitigating unwanted biases with adversarial learning[C]//Proceedings of the 2018 AAAI/ACM Conference on AI, Ethics, and Society. 2018: 335-340.
[18] HARDT M, PRICE E, SREBRO N. Equality of opportunity in supervised learning[C]// Advances in neural information processing systems. 2016: 3315-3323.
[19] MENON A K, WILLIAMSON R C. The cost of fairness in binary classification[C]//Conference on Fairness, Accountability and Transparency. PMLR, 2018: 107-118.
[20] RATHORE A, DEV S, PHILLIPS J M, et al. VERB: Visualizing and interpreting bias mitigation techniques for word representations[J]. arXiv preprintarXiv:2104.02797, 2021.
[21] CHEN Y L, JOO J. Understanding and mitigating annotation bias in facial expression recognition[C]//Proceedings of the IEEE/CVF International Conference on Computer Vision (ICCV). 2021: 14980-14991.
[22] WU H L, MA C, MITRA B, et al. Multi-FR: A multi-objective optimization method for achieving two-sided fairness in E-commerce recommendation[J]. arXiv preprintarXiv:2105.02951, 2021.
[23] XU D P, YUAN S H, ZHANG L, et al. Fairgan+: Achieving fair data generation and classification through generative adversarial nets[C]//2019 IEEE International Conference on Big Data (Big Data). IEEE, 2019: 1401-1406.
[24] DWORK C, IMMORLICA N, KALAI A T, et al. Decoupled classifiers for group-fair and efficient machine learning[C]//Conference on Fairness, Accountability and Transparency. PMLR, 2018: 119-133.
[25] BELLAMY R K, DEY K, HIND M, et al. AI Fairness 360: An extensible toolkit for detecting and mitigating algorithmic bias[J]. IBM Journal of Research and Development, 2019, 63(4/5): 4-1.
[26] AHN Y, LIN Y R. Fairsight: Visual analytics for fairness in decision making[J]. IEEE Transactions on Visualization and Computer Graphics, 2019, 26(1): 1086-1095.
[27] SALEIRO P, KUESTER B, HINKSON L, et al. Aequitas: A bias and fairness audit toolkit[J]. arXiv preprintarXiv:1811.05577, 2018.
[28] ZHANG Q Q, LIU J L, ZHANG Z Q, et al. Fairer machine learning through multi-objective evolutionary learning[C]//International Conference on Artificial Neural Networks. Springer, 2021: 111-123.
[29] VERMA S, RUBIN J. Fairness definitions explained[C]//2018 IEEE/ACM International Workshop on Software Fairness (FairWare). 2018: 1-7.
[30] DWORK C, HARDT M, PITASSI T, et al. Fairness through awareness[C]//Proceedings of the 3rd Innovations in Theoretical Computer Science Conference. 2012: 214-226.
[31] KOHAVI R, et al. Scaling up the accuracy of naive-bayes classifiers: A decision-tree hybrid. [C]//KDD’96: Proceedings of the Second International Conference on Knowledge Discovery and Data Mining. AAAI Press, 1996: 202–207.
[32] BERK R, HEIDARI H, JABBARI S, et al. Fairness in criminal justice risk assessments: The state of the art[J]. Sociological Methods & Research, 2021, 50(1): 3-44.
[33] CHOULDECHOVA A. Fair prediction with disparate impact: A study of bias in recidivism prediction instruments[J]. Big Data, 2017, 5(2): 153-163.
[34] GALHOTRA S, BRUN Y, MELIOU A. Fairness testing: Testing software for discrimination [C]//Proceedings of the 2017 11th Joint Meeting on Foundations of Software Engineering. 2017: 498-510.
[35] SPEICHER T, HEIDARI H, GRGIC-HLACA N, et al. A unified approach to quantifying algorithmic unfairness: Measuring individual &group unfairness via inequality indices[C]// Proceedings of the 24th ACM SIGKDD International Conference on Knowledge Discovery & Data Mining. 2018: 2239-2248.
[36] WAN M Y, ZHA D C, LIU N H, et al. In-processing modeling techniques for machine learning Fairness: A Survey[J]. ACM Transactions on Knowledge Discovery from Data (TKDD), 2022.
[37] KAMISHIMA T, AKAHO S, SAKUMA J. Fairness-aware learning through regularization approach[C]//2011 IEEE 11th International Conference on Data Mining Workshops. IEEE, 2011: 643-650.
[38] DI STEFANO P G, HICKEY J M, VASILEIOU V. Counterfactual fairness: Removing direct effects through regularization[J]. arXiv preprintarXiv:2002.10774, 2020.
[39] KUSNER M, LOFTUS J, RUSSELL C, et al. Counterfactual fairness[C]//NIPS’17: Proceedings of the 31st International Conference on Neural Information Processing Systems. Long Beach, California, USA: Curran Associates Inc., 2017: 4069–4079.
[40] ZAFAR M B, VALERA I, RODRIGUEZ M, et al. Fairness constraints: A flexible approach for fair classification[J]. The Journal of Machine Learning Research, 2019, 20(1): 2737-2778.
[41] ZAFAR M B, VALERA I, ROGRIGUEZ M G, et al. Fairness constraints: Mechanisms for fair Classification[C]//SINGH A, ZHU J. Proceedings of Machine Learning Research: volume 54 Proceedings of the 20th International Conference on Artificial Intelligence and Statistics. PMLR, 2017: 962-970.
[42] WOODWORTH B, GUNASEKAR S, OHANNESSIAN M I, et al. Learning non-discriminatory predictors[C]//KALE S, SHAMIR O. Proceedings of Machine Learning Research: volume 65 Proceedings of the 2017 Conference on Learning Theory. PMLR, 2017: 1920-1953.
[43] COTTER A, JIANG H, GUPTA M, et al. Optimization with non-differentiable constraints with applications to fairness, recall, churn, and other goals[J]. The Journal of Machine Learning Research, 2019, 20(172): 1-59.
[44] MADRAS D, CREAGER E, PITASSI T, et al. Learning adversarially fair and transferable representations[C]//International Conference on Machine Learning. PMLR, 2018: 3384-3393.
[45] BEUTEL A, CHEN J L, ZHAO Z, et al. Data decisions and theoretical implications when adversarially learning fair representations[J]. arXiv preprintarXiv:1707.00075, 2017.
[46] ADEL T, VALERA I, GHAHRAMANI Z, et al. One-network adversarial fairness[C]// Proceedings of the AAAI Conference on Artificial Intelligence: volume 33. 2019: 2412-2420.
[47] LI X X, CUI Z T, WU Y F, et al. Estimating and improving fairness with adversarial learning [J]. arXiv preprintarXiv:2103.04243, 2021.
[48] ZHANG Q Q, LIU J L, ZHANG Z Q, et al. Mitigating unfairness via evolutionary multi-objective ensemble learning[J]. IEEE Transactions on Evolutionary Computation, 2022, Early Access: 1-15.
[49] LIU S Y, VICENTE L N. Accuracy and fairness trade-offs in machine learning: A stochastic multi-objective approach[J]. Computational Management Science, 2022: 1-25.
[50] LIU S Y, VICENTE L N. The stochastic multi-gradient algorithm for multi-objective optimization and its application to supervised machine learning[J]. arXiv preprintarXiv:1907.04472, 2019.
[51] YAO X. Evolving artificial neural networks[J]. Proceedings of the IEEE, 1999, 87(9): 1423- 1447.
[52] GUPTA M R, COTTER A, FARD M M, et al. Proxy fairness[J]. arXiv preprint- arXiv:1806.11212, 2018.
[53] LI M Q, YAO X. Quality evaluation of solution sets in multiobjective optimisation: A survey [J]. ACM Computing Surveys (CSUR), 2019, 52(2): 1-38.
[54] DEB K, PRATAP A, AGARWAL S, et al. A fast and elitist multiobjective genetic algorithm: NSGA-II[J]. IEEE Transactions on Evolutionary Computation, 2002, 6(2): 182-197.
[55] WANG H D, JIAO L C, YAO X. Two_Arch2: An improved two-archive algorithm for many-objective optimization[J]. IEEE Transactions on Evolutionary Computation, 2014, 19(4): 524- 541.
[56] LI B, TANG K, LI J, et al. Stochastic ranking algorithm for many-objective optimization based on multiple indicators[J]. IEEE Transactions on Evolutionary Computation, 2016, 20(6): 924- 938.
[57] MARLER R T, ARORA J S. Survey of multi-objective optimization methods for engineering [J]. Structural and Multidisciplinary Optimization, 2004, 26: 369-395.
[58] MEI Y, NGUYEN S, XUE B, et al. An efficient feature selection algorithm for evolving job shop scheduling rules with genetic programming[J]. IEEE Transactions on Emerging Topics in Computational Intelligence, 2017, 1(5): 339-353.
[59] GONG Z C, CHEN H H, YUAN B, et al. Multiobjective learning in the model space for time series classification[J]. IEEE Transactions on Cybernetics, 2019, 49(3): 918-932.
[60] MINKU L L, YAO X. Software effort estimation as a multiobjective learning problem[J]. ACM Transactions on Software Engineering and Methodology (TOSEM), 2013, 22(4): 1-32.
[61] RUNARSSON T, YAO X. Stochastic ranking for constrained evolutionary optimization[J]. IEEE Transactions on Evolutionary Computation, 2000, 4(3): 284-294.
[62] PESSACH D, SHMUELI E. Algorithmic fairness[J]. arXiv preprintarXiv:2001.09784, 2020.
[63] FRIEDLER S A, SCHEIDEGGER C, VENKATASUBRAMANIAN S, et al. A comparative study of fairness-enhancing interventions in machine learning[C]//Proceedings of the Conference on Fairness, Accountability, and Transparency. 2019: 329-338.
[64] KINGMA D P, BA J. Adam: A method for stochastic optimization[J]. arXiv preprint- arXiv:1412.6980, 2014.
[65] ZHANG Q F, LI H. MOEA/D: A multiobjective evolutionary algorithm based on decomposition [J]. IEEE Transactions on Evolutionary Computation, 2007, 11(6): 712-731.
[66] ZITZLER E, LAUMANNS M, THIELE L. SPEA2: Improving the strength pareto evolutionary algorithm[J]. Technical Report Gloriastrasse, 2001.
[67] TIAN Y, CHENG R, ZHANG X Y, et al. Diversity assessment of multi-objective evolutionary algorithms: Performance Metric and Benchmark Problems[J]. IEEE Computational Intelligence Magazine, 2019, 14(3): 61-74.
[68] PASZKE A, GROSS S, MASSA F, et al. Pytorch: An imperative style, high-performance deep learning library[J]. Advances in Neural Information Processing Systems, 2019, 32.
[69] WATKINS E A, MCKENNA M, CHEN J. The four-fifths rule is not disparate impact: A woeful tale of epistemic trespassing in algorithmic fairness[J]. arXiv preprintarXiv:2202.09519, 2022.
[70] YANG T, LINDER J, BOLCHINI D. DEEP: Design-oriented evaluation of perceived usability [J]. International Journal of Human-Computer Interaction, 2012, 28(5): 308-346.