Study shows label errors impact model disparity metrics, proposing mitigation methods.
problem Impact of label errors on model disparity metrics.
method Empirical study, characterizing label error effects; proposing estimation and relabeling methods.
result Label errors significantly affect model disparity metrics, particularly for minority groups.
Study evaluates when splitting classifiers can improve performance despite disparate treatment.
problem Impact of disparate treatment in classification models.
method Comparison of split classifiers and group-blind classifiers, quantifying performance improvement.
result Proves an equivalent expression for the benefit-of-splitting which can be efficiently computed.
When the performance of a machine learning model varies over groups defined by sensitive attributes (e.g., gender or ethnicity), the performance disparity can be expressed in terms of the probability distributions of the input and output variables over each group. In this paper, we exploit this fact to reduce the dispa…
A new CVaR test reduces group performance disparity detection complexity.
problem Detecting performance disparities across multiple sensitive groups in ML models.
method Conditional Value-at-Risk (CVaR) testing to reduce sample complexity.
result Sample complexity reduced exponentially to be at most the square root of the number of groups.
The paper introduces return parity for fairness in MDPs, addressing delayed and adverse effects.
problem Fairness in MDPs for dynamic domains with delayed and adverse effects.
method Proposes return parity, decomposes return disparity, and develops algorithms for state visitation distributional alignment.
result The proposed algorithms can successfully close the disparity gap while maintaining policy performance.
Develops methods for fair classification under linear disparity constraints.
problem Disparate impacts of machine learning algorithms on protected groups.
method Bayes-optimal fair classification methods via pre-, in-, and post-processing.
result Explicit forms of Bayes-optimal fair classifiers under linear disparity measures.
Following related work in law and policy, two notions of disparity have come to shape the study of fairness in algorithmic decision-making. Algorithms exhibit treatment disparity if they formally treat members of protected subgroups differently; algorithms exhibit impact disparity when outcomes differ across subgroups,…
Selective classification can worsen accuracy disparities between groups.
problem Selective classification can magnify existing accuracy disparities between various groups.
method Study of margin distribution and distributionally-robust models.
result Selective classification can uniformly improve each group on distributionally-robust models.
Paper explores fair classification with bounded disparity using finite datasets.
problem Ensuring fairness in binary classification with protected groups.
method Minimax optimal approach with fairness constraints and demographic disparity control.
result Proposes FairBayes-DDP+ method that achieves minimax lower bound on fairness-aware excess risk.
Bayesian model identifies health disparities in disease progression.
problem Health disparities bias disease progression models.
method Interpretable Bayesian model accounting for three disparities.
result Model identifies and corrects for health disparities.
New framework reduces strategic manipulation cost for minority groups in fair classification.
problem Strategic manipulation disparities in fair classification.
method Constrained optimization framework that constructs classifiers to reduce strategic manipulation cost for minority groups.
result Empirically, the approach reduces strategic manipulation cost for minority groups over multiple real-world datasets.
New method reduces privacy impact on model accuracy for underrepresented groups.
problem Privacy mechanisms disproportionately affect underrepresented groups in machine learning models.
method Proposes DPSGD-F, a modified DPSGD that adjusts group contributions based on clipping bias.
result DPSGD-F removes disparate impact of differential privacy on model accuracy for protected groups.
Develops a new criterion for subgroup fairness in algorithmic decision support.
problem Identifying fair recommendations in algorithms despite group-level differences.
method IJDI criterion and IJDI-Scan approach to detect and mitigate disparities.
result Identifies significant disparities in recommendations across subpopulations.
New method evaluates multiple social disparities using machine learning.
problem Reduction of educational disparities across multiple dimensions.
method Triply-Robust Machine Learning Approach for Causal Decomposition Analysis.
result Simultaneous interventions across multiple domains reduce disparities.
Where machine-learned predictive risk scores inform high-stakes decisions, such as bail and sentencing in criminal justice, fairness has been a serious concern. Recent work has characterized the disparate impact that such risk scores can have when used for a binary classification task. This may not account, however, fo…
Study decomposes racial healthcare disparities via shifts in mediator distributions.
problem Racial disparities in healthcare expenditures and their underlying drivers.
method Framework decomposing disparities into mediator distribution shifts and residual components, using MEPS data.
result Substantial disparities persist even when mediators are equalized, suggesting unmeasured or structural factors.
We provide the asymptotic distribution of the major indexes used in the statistical literature to quantify disparate treatment in machine learning. We aim at promoting the use of confidence intervals when testing the so-called group disparate impact. We illustrate on some examples the importance of using confidence int…
Automated data-driven decision making systems are increasingly being used to assist, or even replace humans in many settings. These systems function by learning from historical decisions, often taken by humans. In order to maximize the utility of these systems (or, classifiers), their training involves minimizing the e…
Conformal prediction sets can lead to unfair outcomes.
problem Disparate impact in decision-making with conformal prediction sets.
method Experiments with human participants to demonstrate disparate impact and propose equalizing set sizes across groups.
result Providing prediction sets that satisfy Equalized Coverage increases disparate impact compared to marginal coverage.
We mathematically compare four competing definitions of group-level nondiscrimination: demographic parity, equalized odds, predictive parity, and calibration. Using the theoretical framework of Friedler et al., we study the properties of each definition under various worldviews, which are assumptions about how, if at a…
Adversarial training can lead to unfair accuracy disparities between different groups.
problem Adversarial training algorithms introduce unfair accuracy disparities between different groups of data.
method Propose a Fair-Robust-Learning (FRL) framework to mitigate unfairness in adversarial defenses.
result Empirical and theoretical validation of FRL's effectiveness in mitigating unfairness.
End-to-end framework learns precise disparity for activity recognition.
problem Precise portrayal of intraclass disparity in activity recognition.
method Knowledge-directed adversarial learning framework with two competitive encoding distributions.
result Robust and generalizable performance on HAR benchmark datasets.
Paper proposes a new algorithm to minimize AUC disparities in machine learning models.
problem Minimizing unfairness in AUC scores for machine learning models.
method Proposes a minimax learning and bias mitigation framework for AUC optimization.
result Proves the convergence of the proposed algorithm to minimize group-level AUC.
Study examines impact of fairness penalties on clinical risk prediction models.
problem Widespread health disparities in machine learning-guided clinical decision-making.
method Empirical study across multiple databases, outcomes, and sensitive attributes.
result Penalizing fairness violations nearly universally degrades model performance and fairness metrics.
Paper tackles fairness in CCA by minimizing correlation disparity error.
problem Fairness issues in CCA.
method Framework to minimize correlation disparity error in CCA.
result Reduces correlation disparity error without sacrificing CCA accuracy.
This paper tackles fair Bayes-optimal classifiers under predictive parity, proving their limitations and proposing a new algorithm.
problem Ensuring fair Bayes-optimal classifiers under predictive parity, especially when group performance levels vary widely.
method Proving the limitations of fair Bayes-optimal classifiers under predictive parity and proposing a new adaptive thresholding algorithm, FairBayes-DPP.
result Fair Bayes-optimal classifiers under predictive parity may not hold if group performance levels vary widely, leading to within-group unfairness.
In the context of machine learning, disparate impact refers to a form of systematic discrimination whereby the output distribution of a model depends on the value of a sensitive attribute (e.g., race or gender). In this paper, we propose an information-theoretic framework to analyze the disparate impact of a binary cla…
Study federates measurement of demographic disparities from quantile sketches.
problem Misalignment of fairness goals with siloed data collection and privacy regulations.
method Federated auditing of demographic parity through score distributions, using Wasserstein--Frechet variance and quantile summaries.
result Proposes a one-shot, communication-efficient protocol to estimate global disparity and its decomposition.
Machine learning models (e.g., speech recognizers) are usually trained to minimize average loss, which results in representation disparity---minority groups (e.g., non-native speakers) contribute less to the training objective and thus tend to suffer higher loss. Worse, as model accuracy affects user retention, a minor…
Study highlights fairness issues in travel behavior prediction models.
problem Ethical challenges in travel behavior analysis using machine learning.
method Operationalized computational fairness by equality of opportunity; compared DNN and DCM; introduced absolute correlation regularization.
result Both DNN and DCM predict disparities across social groups, with DNN outperforming DCM in prediction disparities.
Ensembling DNNs improves minority group performance, leading to fairness.
problem Improving subgroup performances in DNN classifiers.
method Simple homogeneous ensembling of DNNs.
result Fairness naturally emerges from ensembling, improving minority group performance.
EL framework certifies and flags bias in ML models without distributional assumptions.
problem Systematic performance disparities across sensitive subpopulations in ML models.
method Empirical likelihood-based approach for non-parametric fairness auditing.
result EL framework outperforms bootstrap methods in certification and subpopulation discovery.
Whereas previous post-processing approaches for increasing the fairness of predictions of biased classifiers address only group fairness, we propose a method for increasing both individual and group fairness. Our novel framework includes an individual bias detector used to prioritize data samples in a bias mitigation a…
Machine Learning (ML) models trained on data from multiple demographic groups can inherit representation disparity (Hashimoto et al., 2018) that may exist in the data: the model may be less favorable to groups contributing less to the training process; this in turn can degrade population retention in these groups over …
DCEM algorithm reduces bias in machine learning models trained on selective labels.
problem Bias in machine learning models trained on selective labels.
method Disparate Censorship Expectation-Maximization (DCEM) algorithm.
result DCEM improves bias mitigation without sacrificing discriminative performance.
Proposes a method to quantify and decompose disparity in ML models, separating exempt and non-exempt components.
problem Quantifying disparity in ML models, especially when certain features are exempted due to their critical importance.
method Information-theoretic decomposition into exempt and non-exempt components, satisfying desirable properties.
result Proposes a measure of non-exempt disparity that satisfies all desirable properties, and shows impossibility results for observational measures.
What does it mean for an algorithm to be biased? In U.S. law, unintentional bias is encoded via disparate impact, which occurs when a selection process has widely different outcomes for different groups, even as it appears to be neutral. This legal determination hinges on a definition of a protected class (ethnicity, g…
FAIRIF improves fairness in deep learning models without changing the model architecture.
problem Improving fairness in deep learning models trained on sensitive data.
method Two-stage training algorithm that minimizes loss over a reweighted data set, balancing model performance across demographic groups.
result FAIRIF reduces disparity among different groups in classification settings, improving fairness-utility trade-offs.
The increasing impact of algorithmic decisions on people's lives compels us to scrutinize their fairness and, in particular, the disparate impacts that ostensibly-color-blind algorithms can have on different groups. Examples include credit decisioning, hiring, advertising, criminal justice, personalized medicine, and t…
Recommender systems are personalized: we expect the results given to a particular user to reflect that user's preferences. Some researchers have studied the notion of calibration, how well recommendations match users' stated preferences, and bias disparity the extent to which mis-calibration affects different user grou…
Selective regression allows abstention to improve fairness criteria.
problem Selective regression can exacerbate disparities between subgroups.
method Proposes new fairness criteria and two approaches to mitigate performance disparity.
result Proposed fairness criteria ensures performance improvement for every subgroup with reduced coverage.
Unified framework for Bayes-optimal classifiers under group fairness.
problem Mitigating disparate impacts from algorithmic predictions in high-stakes decision-making.
method Unified framework based on Neyman-Pearson argument for deriving Bayes-optimal classifiers under group fairness constraints.
result Proposes FairBayes method that directly controls disparity and achieves optimal fairness-accuracy tradeoff.
Semi-supervised learning benefits the rich more than the poor, affecting fairness.
problem Disparate impact of semi-supervised learning on different sub-populations.
method Theoretical and empirical analysis of a broad family of SSL algorithms using pseudo-labels.
result Semi-supervised learning benefits the rich more than the poor, potentially violating fairness.
The paper examines how adversarial robustness affects accuracy disparity across different classes.
problem Understanding the impact of adversarial robustness on accuracy disparity across different classes.
method Linear classifiers under a Gaussian mixture model, decomposing the impact into inherent and imbalance effects.
result Adversarial robustness consistently degrades standard accuracy in balanced classes, but the class imbalance ratio plays a different role in accuracy disparity.
Study reveals class disparities in balanced datasets through spectral imbalance.
problem Class disparities in balanced datasets are overlooked despite model performance gaps.
method Developed a theoretical framework and studied 11 encoders to diagnose spectral imbalance.
result Identified spectral imbalance as a source of class disparities in balanced datasets.
Standardized fairness measures for continuous risk scores using Wasserstein distance.
problem Quantifying and interpreting group disparities in continuous risk scores.
method Proposes standardized fairness measures based on Wasserstein distance for continuous scores.
result Proposed measures outperform ROC-based fairness measures by being more explicit and quantifying significant biases.
This paper tackles fairness in PCA by balancing it with reconstruction error.
problem Fairness concerns in PCA due to different group representation errors.
method A multi-objective optimization approach to balance fairness and reconstruction error.
result Achieving fairness with minimal loss in reconstruction error.
ROME improves algorithmic fairness by learning latent group structure robustly.
problem Latent subgroup disparities and distribution shifts in machine learning models.
method ROME uses an Expectation-Maximization algorithm for linear models and a neural Mixture-of-Experts for nonlinear settings.
result ROME significantly improves fairness compared to standard methods while maintaining average performance.