Calibration for the (Computationally-Identifiable) Masses

As algorithms increasingly inform and influence decisions made about individuals, it becomes increasingly important to address concerns that these algorithms might be discriminatory. The output of an algorithm can be discriminatory for many reasons, most notably: (1) the data used to train the algorithm might be biased (in various ways) to favor certain populations over others; (2) the analysis of this training data might inadvertently or maliciously introduce biases that are not borne out in the data. This work focuses on the latter concern. We develop and study multicalbration -- a new measure of algorithmic fairness that aims to mitigate concerns about discrimination that is introduced in the process of learning a predictor from data. Multicalibration guarantees accurate (calibrated) predictions for every subpopulation that can be identified within a specified class of computations. We think of the class as being quite rich; in particular, it can contain many overlapping subgroups of a protected group. We show that in many settings this strong notion of protection from discrimination is both attainable and aligned with the goal of obtaining accurate predictions. Along the way, we present new algorithms for learning a multicalibrated predictor, study the computational complexity of this task, and draw new connections to computational learning models such as agnostic learning.

[1]  Leslie G. Valiant,et al.  A theory of the learnable , 1984, STOC '84.

[2]  Silvio Micali,et al.  How to construct random functions , 1986, JACM.

[3]  David Haussler,et al.  Decision Theoretic Generalizations of the PAC Model for Neural Net and Other Learning Applications , 1992, Inf. Comput..

[4]  Michael Kearns,et al.  Efficient noise-tolerant learning from statistical queries , 1993, STOC.

[5]  Umesh V. Vazirani,et al.  An Introduction to Computational Learning Theory , 1994 .

[6]  Shai Ben-David,et al.  Agnostic Boosting , 2001, COLT/EuroCOLT.

[7]  Alvaro Sandroni,et al.  Calibration with Many Checking Rules , 2003, Math. Oper. Res..

[8]  R. Schapire,et al.  Toward efficient agnostic learning , 1992, COLT '92.

[9]  Oded Regev,et al.  On lattices, learning with errors, random linear codes, and cryptography , 2005, STOC '05.

[10]  Yishay Mansour,et al.  From External to Internal Regret , 2005, J. Mach. Learn. Res..

[11]  Adam Tauman Kalai,et al.  On agnostic boosting and parity learning , 2008, STOC.

[12]  Subhash Khot,et al.  Minimizing Wide Range Regret with Time Selection Functions , 2008, COLT.

[13]  Adam Tauman Kalai,et al.  Potential-Based Agnostic Boosting , 2009, NIPS.

[14]  Madhur Tulsiani,et al.  Regularity, Boosting, and Efficiently Simulating Every High-Entropy Distribution , 2009, 2009 24th Annual IEEE Conference on Computational Complexity.

[15]  Vitaly Feldman,et al.  Distribution-Specific Agnostic Boosting , 2009, ICS.

[16]  Guy N. Rothblum,et al.  A Multiplicative Weights Mechanism for Privacy-Preserving Data Analysis , 2010, 2010 IEEE 51st Annual Symposium on Foundations of Computer Science.

[17]  Aaron Roth,et al.  Iterative Constructions and Private Data Release , 2011, TCC.

[18]  Toniann Pitassi,et al.  Fairness through awareness , 2011, ITCS '12.

[19]  Shai Shalev-Shwartz,et al.  Online Learning and Online Convex Optimization , 2012, Found. Trends Mach. Learn..

[20]  Aaron Roth,et al.  The Algorithmic Foundations of Differential Privacy , 2014, Found. Trends Theor. Comput. Sci..

[21]  Tim Roughgarden,et al.  Privately Solving Linear Programs , 2014, ICALP.

[22]  Avrim Blum,et al.  The Ladder: A Reliable Leaderboard for Machine Learning Competitions , 2015, ICML.

[23]  Toniann Pitassi,et al.  Preserving Statistical Validity in Adaptive Data Analysis , 2014, STOC.

[24]  Toniann Pitassi,et al.  Generalization in Adaptive Data Analysis and Holdout Reuse , 2015, NIPS.

[25]  Percy Liang,et al.  Calibrated Structured Prediction , 2015, NIPS.

[26]  Toniann Pitassi,et al.  The reusable holdout: Preserving validity in adaptive data analysis , 2015, Science.

[27]  Jonathan Ullman,et al.  Private Multiplicative Weights Beyond Linear Queries , 2014, PODS.

[28]  Raef Bassily,et al.  Algorithmic stability for adaptive data analysis , 2015, STOC.

[29]  Aaron Roth,et al.  Fairness in Learning: Classic and Contextual Bandits , 2016, NIPS.

[30]  Seth Neel,et al.  Fair Algorithms for Infinite and Contextual Bandits , 2016, 1610.09559.

[31]  Nathan Srebro,et al.  Equality of Opportunity in Supervised Learning , 2016, NIPS.

[32]  Elad Hazan,et al.  Introduction to Online Convex Optimization , 2016, Found. Trends Optim..

[33]  Benjamin Fish,et al.  A Confidence-Based Approach for Balancing Fairness and Accuracy , 2016, SDM.

[34]  Alexandra Chouldechova,et al.  Fair prediction with disparate impact: A study of bias in recidivism prediction instruments , 2016, Big Data.

[35]  Alexandra Chouldechova,et al.  Fairer and more accurate, but for whom? , 2017, ArXiv.

[36]  Yang Liu,et al.  Calibrated Fairness in Bandits , 2017, ArXiv.

[37]  Blake E. Woodworth,et al.  Lower Bound for Randomized First Order Convex Optimization , 2017, 1709.03594.

[38]  Seth Neel,et al.  A Convex Framework for Fair Regression , 2017, ArXiv.

[39]  Algorithmic decision making and the cost of fairness , 2017, 1701.08230.

[40]  Alon Rosen,et al.  Pseudorandom Functions: Three Decades Later , 2017, Tutorials on the Foundations of Cryptography.

[41]  Jon M. Kleinberg,et al.  Inherent Trade-Offs in the Fair Determination of Risk Scores , 2016, ITCS.

[42]  Bernhard Schölkopf,et al.  Avoiding Discrimination through Causal Reasoning , 2017, NIPS.

[43]  Jon M. Kleinberg,et al.  On Fairness and Calibration , 2017, NIPS.

[44]  Seth Neel,et al.  Preventing Fairness Gerrymandering: Auditing and Learning for Subgroup Fairness , 2017, ICML.

[45]  Sergiu Hart,et al.  Smooth calibration, leaky forecasts, finite recall, and Nash dynamics , 2018, Games Econ. Behav..

[46]  Guy N. Rothblum,et al.  Fairness Through Computationally-Bounded Awareness , 2018, NeurIPS.

[47]  Nisheeth K. Vishnoi,et al.  Ranking with Fairness Constraints , 2017, ICALP.