Fairness Constraints Research Articles

Distribution Fairness in Multiplayer AI Using Shapley Constraints

Experience management (EM) agents in multiplayer serious games face unique challenges and responsibilities regarding the fair treatment of players. One such challenge is the Greedy Bandit Problem that arises when using traditional Multi-Armed Bandits (MABs) as EM agents, which results in some players routinely prioritized while others may be ignored. We will show that this problem can be a cause of player non-adherence in a multiplayer serious game played by human users. To mitigate this effect, we propose a new bandit strategy, the Shapley Bandit, which enforces fairness constraints in its treatment of players based on the Shapley Value. We evaluate our approach via simulation with virtual players, finding that the Shapley Bandit can be effective in providing more uniform treatment of players while incurring only a slight cost in overall performance to a typical greedy approach. Our findings highlight the importance of fair treatment among players as a goal of multiplayer EM agents and discuss how addressing this issue may lead to more effective agent operation overall. The study contributes to the understanding of player modeling and EM in serious games and provides a promising approach for balancing fairness and engagement in multiplayer environments.

Group Equality in Adaptive Submodular Maximization

In this paper, we study the classic submodular maximization problem subject to a group equality constraint under both nonadaptive and adaptive settings. It is shown that the utility function of many machine learning applications, including data summarization, influence maximization in social networks, and personalized recommendation, satisfies the property of submodularity. Hence, maximizing a submodular function subject to various constraints can be found at the heart of many of those applications. On a high level, submodular maximization aims to select a group of most representative items (e.g., data points). However, the design of most existing algorithms does not incorporate the fairness constraint, leading to underrepresentation or overrepresentation of some particular groups. This motivates us to study the submodular maximization problem with group equality, in which we aim to select a group of items to maximize a (possibly nonmonotone) submodular utility function subject to a group equality constraint. To this end, we develop the first constant-factor approximation algorithm for this problem. The design of our algorithm is robust enough to be extended to solving the submodular maximization problem under a more complicated adaptive setting. Moreover, we further extend our study to incorporating a global cardinality constraint and other fairness notations. History: Accepted by Andrea Lodi, Area Editor for Design & Analysis of Algorithms–Discrete. Supplemental Material: The online supplement is available at https://doi.org/10.1287/ijoc.2022.0384 .

Enabling Equal Opportunity in Logistic Regression Algorithm

Research Question: This paper aims at adjusting the logistic regression algorithm to mitigate unwanted discrimination shown towards race, gender, etc. Motivation: Decades of research in the field of algorithm design have been dedicated to making a better prediction model. Many algorithms are designed and improved, which made them better than the judgments of people and even experts. However, in recent years it has been discovered that predictive models can make unwanted discrimination. Such unwanted discrimination in the predictive model can lead to legal consequences. In order to mitigate the problem of unwanted discrimination, we propose equal opportunity between privileged and discriminated groups in the logistic regression algorithm. Idea: Our idea is to add a regularization term in the goal function of the logistic regression. Therefore, our predictive model will solve both the social problem and the predictive problem. More specifically, our model will provide fair and accurate predictions. Data: The data used in this research present U.S. census data describing individuals using personal characteristics with a goal to provide a binary classification model for predicting if an individual has an annual salary above $50k. The dataset used is known for disparate impact regarding female individuals. In addition, we used the COMPAS dataset aimed at predicting recidivism. COMPAS is biased toward African-Americans. Tools: We developed a novel regularization technique for equal opportunity in the logistic regression algorithm. The proposed regularization is compared against classical logistic regression and fairness constraint logistic regression, using a ten-fold cross-validation. Findings: The results suggest that equal opportunity logistic regression manages to create a fair prediction model. More specifically, our model improved both disparate impact and equal opportunity compared to classical logistic regression, with a minor loss in prediction accuracy. Compared to the disparate impact constrained logistic regression, our approach has higher prediction accuracy and equal opportunity, while having a lower disparate impact. By inspecting the coefficients of our approach and classical logistic regression, one can see that proxy attribute coefficients are reduced to very low values. Contribution: The main contribution of this paper is in the methodological part. More specifically, we implemented an equal opportunity in the logistic regression algorithm.

Fairness criteria for allocating indivisible chores: connections and efficiencies

We study several fairness notions in allocating indivisible chores (i.e., items with disutilities) to agents who have additive and submodular cost functions. The fairness criteria we are concerned with are envy-free up to any item, envy-free up to one item, maximin share (MMS), and pairwise maximin share (PMMS), which are proposed as relaxations of envy-freeness in the setting of additive cost functions. For allocations under each fairness criterion, we establish their approximation guarantee for other fairness criteria. Under the additive setting, our results show strong connections between these fairness criteria and, at the same time, reveal intrinsic differences between goods allocation and chores allocation. However, such strong relationships cannot be inherited by the submodular setting, under which PMMS and MMS are no longer relaxations of envy-freeness and, even worse, few non-trivial guarantees exist. We also investigate efficiency loss under these fairness constraints and establish their prices of fairness.

FairAW – Additive weighting without discrimination

With growing awareness of the societal impact of decision-making, fairness has become an important issue. More specifically, in many real-world situations, decision-makers can unintentionally discriminate a certain group of individuals based on either inherited or appropriated attributes, such as gender, age, race, or religion. In this paper, we introduce a post-processing technique, called fair additive weighting (FairAW) for achieving group and individual fairness in multi-criteria decision-making methods. The methodology is based on changing the score of an alternative by imposing fair criteria weights. This is achieved through minimization of differences in scores of individuals subject to fairness constraint. The proposed methodology can be successfully used in multi-criteria decision-making methods where the additive weighting is used to evaluate scores of individuals. Moreover, we tested the method both on synthetic and real-world data, and compared it to Disparate Impact Remover and FA*IR methods that are commonly used in achieving fair scoring of individuals. The obtained results showed that FairAW manages to achieve group fairness in terms of statistical parity, while also retaining individual fairness. Additionally, our approach managed to obtain the best equality in scoring between discriminated and privileged groups.

Fair Max-Min Diversity Maximization in Streaming and Sliding-Window Models.

Diversity maximization is a fundamental problem with broad applications in data summarization, web search, and recommender systems. Given a set X of n elements, the problem asks for a subset S of k≪n elements with maximum diversity, as quantified by the dissimilarities among the elements in S. In this paper, we study diversity maximization with fairness constraints in streaming and sliding-window models. Specifically, we focus on the max-min diversity maximization problem, which selects a subset S that maximizes the minimum distance (dissimilarity) between any pair of distinct elements within it. Assuming that the set X is partitioned into m disjoint groups by a specific sensitive attribute, e.g., sex or race, ensuring fairness requires that the selected subset S contains ki elements from each group i∈[m]. Although diversity maximization has been extensively studied, existing algorithms for fair max-min diversity maximization are inefficient for data streams. To address the problem, we first design efficient approximation algorithms for this problem in the (insert-only) streaming model, where data arrive one element at a time, and a solution should be computed based on the elements observed in one pass. Furthermore, we propose approximation algorithms for this problem in the sliding-window model, where only the latest w elements in the stream are considered for computation to capture the recency of the data. Experimental results on real-world and synthetic datasets show that our algorithms provide solutions of comparable quality to the state-of-the-art offline algorithms while running several orders of magnitude faster in the streaming and sliding-window settings.

FairSwiRL: fair semi-supervised classification with representation learning

Semi-supervised learning has shown its potential in many real-world applications where only few labeled examples are available. However, when some fairness constraints need to be satisfied, semi-supervised classification models often struggle as they are required to cope with the lack of sufficient information for predicting the target variable while forgetting its relationships with any sensitive and potentially discriminatory attribute. To address this issue, we propose a fair semi-supervised representation learning architecture that leads to fair and accurate classification results even in very challenging scenarios with few labeled (but biased) instances. We show experimentally that our model can be easily adopted in very general settings, as the learned representations may be employed to train any supervised classifier. Moreover, when applied to several synthetic and real-world datasets, our method is competitive with state-of-the-art fair semi-supervised approaches.

Fair-CDA: Continuous and Directional Augmentation for Group Fairness

In this work, we propose Fair-CDA, a fine-grained data augmentation strategy for imposing fairness constraints. We use a feature disentanglement method to extract the features highly related to the sensitive attributes. Then we show that group fairness can be achieved by regularizing the models on transition paths of sensitive features between groups. By adjusting the perturbation strength in the direction of the paths, our proposed augmentation is controllable and auditable. To alleviate the accuracy degradation caused by fairness constraints, we further introduce a calibrated model to impute labels for the augmented data. Our proposed method does not assume any data generative model and ensures good generalization for both accuracy and fairness. Experimental results show that Fair-CDA consistently outperforms state-of-the-art methods on widely-used benchmarks, e.g., Adult, CelebA and MovieLens. Especially, Fair-CDA obtains an 86.3% relative improvement for fairness while maintaining the accuracy on the Adult dataset. Moreover, we evaluate Fair-CDA in an online recommendation system to demonstrate the effectiveness of our method in terms of accuracy and fairness.

Faster Fair Machine via Transferring Fairness Constraints to Virtual Samples

Faster Fair Machine via Transferring Fairness Constraints to Virtual Samples

Accurate Fairness: Improving Individual Fairness without Trading Accuracy

Accuracy and individual fairness are both crucial for trustworthy machine learning, but these two aspects are often incompatible with each other so that enhancing one aspect may sacrifice the other inevitably with side effects of true bias or false fairness. We propose in this paper a new fairness criterion, accurate fairness, to align individual fairness with accuracy. Informally, it requires the treatments of an individual and the individual's similar counterparts to conform to a uniform target, i.e., the ground truth of the individual. We prove that accurate fairness also implies typical group fairness criteria over a union of similar sub-populations. We then present a Siamese fairness in-processing approach to minimize the accuracy and fairness losses of a machine learning model under the accurate fairness constraints. To the best of our knowledge, this is the first time that a Siamese approach is adapted for bias mitigation. We also propose fairness confusion matrix-based metrics, fair-precision, fair-recall, and fair-F1 score, to quantify a trade-off between accuracy and individual fairness. Comparative case studies with popular fairness datasets show that our Siamese fairness approach can achieve on average 1.02%-8.78% higher individual fairness (in terms of fairness through awareness) and 8.38%-13.69% higher accuracy, as well as 10.09%-20.57% higher true fair rate, and 5.43%-10.01% higher fair-F1 score, than the state-of-the-art bias mitigation techniques. This demonstrates that our Siamese fairness approach can indeed improve individual fairness without trading accuracy. Finally, the accurate fairness criterion and Siamese fairness approach are applied to mitigate the possible service discrimination with a real Ctrip dataset, by on average fairly serving 112.33% more customers (specifically, 81.29% more customers in an accurately fair way) than baseline models.

Social Bias Meets Data Bias: The Impacts of Labeling and Measurement Errors on Fairness Criteria

Although many fairness criteria have been proposed to ensure that machine learning algorithms do not exhibit or amplify our existing social biases, these algorithms are trained on datasets that can themselves be statistically biased. In this paper, we investigate the robustness of existing (demographic) fairness criteria when the algorithm is trained on biased data. We consider two forms of dataset bias: errors by prior decision makers in the labeling process, and errors in the measurement of the features of disadvantaged individuals. We analytically show that some constraints (such as Demographic Parity) can remain robust when facing certain statistical biases, while others (such as Equalized Odds) are significantly violated if trained on biased data. We provide numerical experiments based on three real-world datasets (the FICO, Adult, and German credit score datasets) supporting our analytical findings. While fairness criteria are primarily chosen under normative considerations in practice, our results show that naively applying a fairness constraint can lead to not only a loss in utility for the decision maker, but more severe unfairness when data bias exists. Thus, understanding how fairness criteria react to different forms of data bias presents a critical guideline for choosing among existing fairness criteria, or for proposing new criteria, when available datasets may be biased.

Comfort violation minimization oriented energy management strategy for a renewable-assisted stand-alone multi-energy electric vehicle service station

In this study, a comfort violation minimization oriented energy management strategy is proposed for a stand-alone electric vehicle (EV) service station serving multiple EV types in a multi-energy manner. The main purpose of this strategy is to minimize the discomfort of EV owners due to insufficient energy supply while considering the fairness among users and optimizing the utilization of renewable energy sources. The criteria for selecting this strategy include: (1) mitigating customer dissatisfaction arising from inadequate energy supply, (2) ensuring fairness in energy distribution among different EV types, and (3) maximizing the use of renewable power sources, such as photovoltaic (PV) units. The developed mixed-integer linear programming (MILP) model comprises a PV unit as the main power source, with an energy storage system (ESS) and an electrolyzer-supported main hydrogen storage unit fulfilling the energy and hydrogen demands of battery electric vehicles (BEVs) and fuel cell electric vehicles (FCEVs), respectively. A comfort violation fairness constraint is also introduced to guarantee a fair service for all EV owners, thereby enhancing the social acceptability of the provided EV service.Simulation results, obtained under various case studies considering component size changes and comfort fairness constraint relaxation, demonstrate the importance of striking a balance between investing in larger components and allowing for a certain level of customer dissatisfaction. The analysis revealed that relaxation of the fairness constraint led to a reduction in overall comfort violation by up to 15 % for both BEVs and FCEVs combined; however, this relaxation resulted in significant differences in comfort violation among different EV owner types, with variations ranging from a minimum of 0 % to a maximum of >90 %. These findings emphasize the significance of considering fairness in energy distribution to prevent potential dissatisfaction among end-users. Future studies will explore the development of a grid-connected system and compare the stand-alone structure in a techno-economic context, considering the possible impact of different EV types on grid operation during normal and resiliency-oriented conditions.

Beyond submodularity: a unified framework of randomized set selection with group fairness constraints

Machine learning algorithms play an important role in a variety of important decision-making processes, including targeted advertisement displays, home loan approvals, and criminal behavior predictions. Given the far-reaching impact of these algorithms, it is crucial that they operate fairly, free from bias or prejudice towards certain groups in the population. Ensuring impartiality in these algorithms is essential for promoting equality and avoiding discrimination. To this end we introduce a unified framework for randomized subset selection that incorporates group fairness constraints. Our problem involves a global utility function and a set of group utility functions for each group, here a group refers to a group of individuals (e.g., people) sharing the same attributes (e.g., gender). Our aim is to generate a distribution across feasible subsets, specifying the selection probability of each feasible set, to maximize the global utility function while meeting a predetermined quota for each group utility function in expectation. Note that there may not necessarily be any direct connections between the global utility function and each group utility function. We demonstrate that this framework unifies and generalizes many significant applications in machine learning and operations research. Our algorithmic results either improves the best known result or provide the first approximation algorithms for new applications.

Energy-efficient data transmission with proportional rate fairness for NANs of smart grid communication network

In smart grid, the neighborhood area network (NAN) serves as a bridge connecting smart meters and meter data management system (MDMS) and is one of the most important components of the smart grid communication network. In this work, we study the energy-efficient data transmissions from NAN gateways (i.e., routers) to the concentrator that connects the MDMS. Particularly, to assure that each router associated with diverse number of associated smart meters can achieve a required data rate, a set of proportional rate fairness constraints are imposed into the optimization problem. In this case, an optimization problem that balances energy efficiency and fairness is formulated. Due to the fractional form of objective function, the combinatorial constraints on channel allocation variables, and the proportional rate fairness requirements, the formulated energy-efficient data transmission problem is non-convex and extremely computationally complex. In order to solve the problem efficiently, firstly, a subtractive transformation is introduced to handle the objective function, and an iterative algorithm based on Dinkelbach method is proposed; then for the inner loop optimization problem in each iteration, a low-complexity suboptimal algorithm that separates channel allocation and power distribution is developed. Finally, numerical results demonstrate that the proposed algorithm converges in a few steps and achieves a near-optimal energy efficiency performance.

Detox Loss: Fairness Constraints for Learning With Imbalanced Data

Recent studies have highlighted a major caveat in several popular deep learning based face recognition algorithms. Despite their high accuracy, they have shown to exhibit biased behavior by achieving sub-par performance on images belonging to different sub-groups. Such behavior can often lead to unfair predictions, thus presenting a need to develop fairer models with unprejudiced behavior. To this effect, this research proposes Detox loss, a novel bias invariant feature learning loss function for learning unbiased models. The proposed loss can be used to: (i) learn fairer deep learning classifiers, and (ii) mitigate bias from existing pre-trained networks, especially in the challenging constraint of imbalanced training data with respect to a protected attribute. Conceptually, the Detox loss enforces that the learned features are distinguished based on the task label only, while eliminating any distinction based on the biasing-attribute. This is achieved by incorporating three fairness constraints while training with the traditional classification loss: (i) proposed Exclusion loss, (ii) proposed Inclusion loss, and (iii) proposed Feature-distillation loss. The efficacy of the Detox loss is demonstrated on two facial analysis tasks: (i) age-group prediction and (ii) gender prediction, under the protected attribute of ethnicity for learning fairer models and de-biasing existing models, with varying imbalanced training data distributions. Across two different protocols, three setups, and two tasks, the Detox loss obtains state of the art performance without the need of multi-labeled training data. For example, on the challenging Pilot Parliaments Benchmark (PPB) dataset, the Detox loss obtains a balanced accuracy and F1 score of 96.1% and 96.6%, as compared to the state-of-the-art performance of 94.1% and 93.6%, respectively.

Group fairness in non-monotone submodular maximization

Maximizing a submodular function has a wide range of applications in machine learning and data mining. One such application is data summarization whose goal is to select a small set of representative and diverse data items from a large dataset. However, data items might have sensitive attributes such as race or gender, in this setting, it is important to design fairness-aware algorithms to mitigate potential algorithmic bias that may cause over- or under- representation of particular groups. Motivated by that, we propose and study the classic non-monotone submodular maximization problem subject to novel group fairness constraints. Our goal is to select a set of items that maximizes a non-monotone submodular function, while ensuring that the number of selected items from each group is proportionate to its size, to the extent specified by the decision maker. We develop the first constant-factor approximation algorithms for this problem. We also extend the basic model to incorporate an additional global size constraint on the total number of selected items.

Energy efficient federated learning in internet of vehicles: A game theoretic scheme

AbstractA promising alternative to conventional centralized machine learning that protects privacy is federated learning. Energy efficiency, meanwhile, appears as a major problem in federated learning for the Internet of Vehicles due to frequent model aggregations. To reduce the convergence time, existing research focuses on improving vehicular communication. However, it increases the computational and communication load on servers and disregards the true energy costs of vehicles. Thus, we propose an energy‐efficient federated learning scheme based on two‐stage game theories. Specifically, the payoff function first takes into account the energy cost as well as the contribution of each vehicle, which wins the fair payment based on the Shapley value via a cooperative game. Second, under the fairness constraint, energy‐efficient vehicles cooperatively form several coalitions for further federated worker selection through multiple rounds of the hedonic game. Additionally, we also develop a multiagent reinforcement learning algorithm that is leveraged for coalition formation to lower the energy cost in the game. Furthermore, we demonstrate that the game in our scheme is Nash stable. Extensive experimental results show that the scheme improves energy efficiency by around 68.8% in the best case compared to the most widely used method.

Fairness-Constrained Rate Optimization of Multi-Link Slotted Aloha

This letter optimizes the data rate performance of multi-link slotted Aloha networks, in which Multi-Link Devices (MLDs) coexist with multiple groups of Single-Link Devices (SLDs), where different groups reside in distinct channels. MLDs can transmit/receive packets via multiple channels, causing interference to all groups of SLDs and severe rate unfairness issue. To ensure SLDs in different channels can acquire target ratios of data rate, explicit expressions of sum rate and group data rate are derived, based on which the optimal transmission probabilities of MLDs/SLDs for maximizing the sum rate without or with fairness constraints are obtained. The analysis reveals the coupling effect of the behavior of devices in different channels, the tradeoff between the optimal sum rate performance and the fairness, and the effect of channel fading on benefit of using multi-link transmission capability.

Achieving Multi-Beam Gain in Intelligent Reflecting Surface Assisted Wireless Energy Transfer

Intelligent reflecting surface (IRS) is a promising technology to boost the efficiency of wireless energy transfer (WET) systems. However, for a multiuser WET system, simultaneous multi-beam energy transmission is generally required to achieve the maximum performance, which may not be implemented by using the IRS having only a single set of coefficients. As a result, it remains unknown how to exploit the IRS to approach such a performance upper bound. To answer this question, we aim to maximize the total harvested energy of a multiuser WET system subject to the user fairness constraints and the non-linear energy harvesting model. We first consider the static IRS beamforming scheme, which shows that the optimal IRS reflection matrix obtained by applying semidefinite relaxation is indeed of high rank in general as the number of energy receivers (ERs) increases, due to which the resulting rank-one solution by applying Gaussian Randomization may lead to significant loss. To achieve the multi-beam gain, we then propose a general time-division based novel framework by exploiting the IRS's dynamic passive beamforming. Moreover, it is able to achieve a good balance between the system performance and complexity by controlling the number of IRS shift patterns. Finally, we also propose a time-division multiple access (TDMA) based passive beamforming design for performance comparison. Simulation results demonstrate the necessity of multi-beam transmission and the superiority of the proposed dynamic IRS beamforming scheme over existing schemes.

Lq regularization for fair artificial intelligence robust to covariate shift

AbstractIt is well recognized that historical biases exist in training data against a certain sensitive group (e.g., non‐White, women) which are socially unacceptable, and these unfair biases are inherited in trained artificial intelligence (AI) models. Various learning algorithms have been proposed to remove or alleviate unfair biases in trained AI models. In this paper, we consider another type of bias in training data so‐called covariate shift in view of fair AI. Here, covariate shift means that training data do not represent the population of interest well. Covariate shift occurs when special sampling designs (e.g., stratified sampling) are used when collecting training data, or the population where training data are collected is different from the population of interest. When covariate shift exists, fair AI models on training data may not be fair in test data. To ensure fairness on test data, we develop computationally efficient learning algorithms robust to covariate shifts. In particular, we propose a robust fairness constraint based on the Lq norm which is a generic algorithm to be applied to various fairness AI problems without much hampering. By analyzing multiple benchmark datasets, we show that our proposed robust fairness AI algorithm improves existing fair AI algorithms much in terms of the fairness‐accuracy tradeoff to covariate shift and has significant computational advantages compared to other robust fair AI algorithms.