Cause Of Bias Research Articles

A-210 Ldl-c to non-hdl-c ratio method: a simple way to determine bias for ldl-c methods on hypertriglyceridemic samples

Abstract Background Beta-quantification (BQ) is the reference method for LDL-C but is not widely available. It is difficult, therefore, for clinical laboratories to independently assess the accuracy of their methods for either calculating or directly measuring LDL-C. Our goal was to investigate if an interrelationship between the tests in the standard lipid panel could be used to compare the performance of different LDL-C methods against BQ, specifically in samples with hypertriglyceridemia, a major cause of bias for LDL-C testing. Methods BQ results from Mayo Medical Labs (n=39,667) and the NIH (n=18,338) were used to identify a linear relationship by least-squares regression analysis between lipid panel test results and LDL-C to facilitate a statistical comparison between different LDL-C methods. Results It is well established that the number of VLDL particles increases as TG increases, because VLDL is the main carrier of TG. As a consequence, the ratio of LDL-C to non-HDL-C is inversely related to TG, which can be made linear by taking the square root of TG (Figure). Nearly identical regression equations were found for this relationship, using two different BQ datasets. When LDL-C was calculated by either Sampson equation, the LDL-C/non-HDL-C regression lines nearly overlapped with the one for BQ. In contrast, the Martin equation showed a positive bias with increasing TG, whereas the Friedewald equation showed a negative bias, consistent with past reports on their LDL-C biases with hypertriglyceridemia. In the UKBiobank, the Beckman direct LDL-C showed a positive bias for TG, which was confirmed in CAP proficiency test surveys. By simulation analysis, approximately 80 samples with high (>200 mg/dL) and low (<100 mg/dL) TG is statistically sufficient to assess if the LDL-C/non-HDL-C regressions lines of LDL-C methods differ from BQ. Conclusions The LDL-C/non-HDL-C ratio method is a simple way to examine LDL-C methods for bias on hypertriglyceridemic samples.

Following Spinoza’s conjecture and image: Representativeness and availability as generators of conscious errors in space and time

The article will present an intertextual encounter between Spinoza’s approach and the work of Kahneman and others. on cognitive bias. The discussion will focus on the link between imagination and memory and space and time as well as the former’s association with representativeness, availability, and occasion noise as the causes of bias in both systems of thought. Spinoza determined that the manner of human thought has common traits but also that there is thought which accords with reality alongside thought which is erroneous. Like Spinoza, contemporary theory also seeks to address the causes of such erroneous thought and to offer remedies for its prevention.

Fair AI-powered orthopedic image segmentation: addressing bias and promoting equitable healthcare

AI-powered segmentation of hip and knee bony anatomy has revolutionized orthopedics, transforming pre-operative planning and post-operative assessment. Despite the remarkable advancements in AI algorithms for medical imaging, the potential for biases inherent within these models remains largely unexplored. This study tackles these concerns by thoroughly re-examining AI-driven segmentation for hip and knee bony anatomy. While advanced imaging modalities like CT and MRI offer comprehensive views, plain radiographs (X-rays) predominate the standard initial clinical assessment due to their widespread availability, low cost, and rapid acquisition. Hence, we focused on plain radiographs to ensure the utilization of our contribution in diverse healthcare settings, including those with limited access to advanced imaging technologies. This work provides insights into the underlying causes of biases in AI-based knee and hip image segmentation through an extensive evaluation, presenting targeted mitigation strategies to alleviate biases related to sex, race, and age, using an automatic segmentation that is fair, impartial, and safe in the context of AI. Our contribution can enhance inclusivity, ethical practices, equity, and an unbiased healthcare environment with advanced clinical outcomes, aiding decision-making and osteoarthritis research. Furthermore, we have made all the codes and datasets publicly and freely accessible to promote open scientific research.

Efficient circular repeated measurements designs strongly balanced to control carryover effects

In experiments related to agriculture, psychology, medicine, animal sciences, pharmacology, and other fields, repeated measurements designs (RMDs) are utilized which are economical design. However, the use of these designs may result in carryover effects, which are the primary cause of bias. Balanced or strongly balanced RMDs can control these effects. Strongly balanced RMDs estimate the direct effects and carryover effects independently; therefore, these designs have an edge over the balanced RMDs. In this article, therefore, some general procedures are presented to obtain minimal circular strongly and nearly strongly balanced RMDs in periods of k different sizes with high efficiency of Separability and of carryover effects.

Problems of bias and discrimination of human capital in artificial intelligence systems

The article is devoted to the study of the genesis of understanding the causes of bias and discrimination of human capital from traditional forms to the era of digitalisation and artificial intelligence (hereinafter referred to as AI). In this article, the authors continue their case studies in the field of studying the ethics of AI, advantages and risks of its widespread distribution and use. The main purpose of writing this work is to study the question of how the use of AI and algorithmic solutions aggravates the problem of bias and discrimination of human capital which is inevitably associated with the emergence of AI. The authors of the article present a brief overview of the retrospective, discover new modern forms of discrimination generated by the action of AI and subject them to open discussion, offering their vision of solving issues on neutralisation the risks caused by the use of AI technologies in relation to certain groups of workers. The authors consider in the article the manifestations of bias and discrimination in society in general and in the field of human resource management in particular, identify possible threats of discrimination as a result of the spread of AI and the consequences of these threats.

Understanding Biases in Indian Ocean Seasonal SST in CMIP6 Models

AbstractThe latest generation of climate models continue to exhibit biases in their representation of climatological sea surface temperature (SST), affecting their ability to simulate climate variability including the Indian Ocean Dipole which is typically too strong. Here, we analyze the surface layer heat budget of the Indian Ocean to diagnose the processes leading to biases in climatological SST biases in 20 Coupled Model Intercomparison Project Phase 6 (CMIP6) models, in comparison to a suite of observational and reanalysis products. In the western tropical Indian Ocean, we find that weaker than observed winds reduce the strength of surface currents leading to warm SST bias. In the south‐eastern tropical Indian Ocean, overly strong southeasterly winds are associated with overestimated coastal upwelling that increases cooling across CMIP6 models. In the Arabian Sea, overly strong surface winds increase latent heat loss and leads to cool SST biases. We also analyze other regions like the Bay of Bengal, where a persistent cool bias cannot be explained by seasonal heat budgets, and southern Indian Ocean, where overly strong surface winds are responsible for cool SST biases. These biases in surface processes are supported by intermodel relationships between relevant variables, thus explaining differences in simulating the climatological SSTs across the model ensemble. Our analysis suggests that biases in atmospheric processes in particular surface winds are a primary cause of biases in Indian Ocean SST. Reducing these biases would improve the simulation of climate variability in the Indian Ocean toward more reliable climate projections and predictions.

Construction of efficient classes of circular balanced repeated measurements designs with R

Pharmacology, medicine, psychology, and the animal sciences all employ repeated measurement designs (RMDs). However, RMDs may experience carryover effects, which are the primary cause of bias in treatment effect estimation. In order to eliminate the carryover effects for odd v (the number of treatments), minimal circular balanced and strongly balanced repeated measurement designs (RMDs) are the ones that should be used. The minimal circular partially balanced and weakly balanced RMDs are used for even v. In order to obtain these important classes of minimal circular RMDs in periods of equal, two, and three different sizes, an R-based algorithm is developed in this article. The newly developed algorithm has made so simple the construction of balanced RMDs and their generalized classes. As a result, it is a novel piece of research.

Assessing and Mitigating Bias in Artificial Intelligence: A Review

Abstract: There has been an exponential increase in discussions about bias in Artificial Intelligence (AI) systems. Bias in AI has typically been defined as a divergence from standard statistical patterns in the output of an AI model, which could be due to a biased dataset or biased assumptions. While the bias in artificially taught models is attributed able to bias in the dataset provided by humans, there is still room for advancement in terms of bias mitigation in AI models. The failure to detect bias in datasets or models stems from the "black box" problem or a lack of understanding of algorithmic outcomes. This paper provides a comprehensive review of the analysis of the approaches provided by researchers and scholars to mitigate AI bias and investigate the several methods of employing a responsible AI model for decision-making processes. We clarify what bias means to different people, as well as provide the actual definition of bias in AI systems. In addition, the paper discussed the causes of bias in AI systems thereby permitting researchers to focus their efforts on minimising the causes and mitigating bias. Finally, we recommend the best direction for future research to ensure the discovery of the most accurate method for reducing bias in algorithms. We hope that this study will help researchers to think from different perspectives while developing unbiased systems.

Teaching Structural Competency in Law School: Interdisciplinary Inspiration from Medical Legal Partnerships and Health-Related Disciplines to Meet ABA Standard 303(c).

Law Schools are now required to provide education to law students on bias, cross-cultural competency, and racism under ABA Standard 303(c). Law clinics, with their social justice orientation, have long taught about structural causes of bias and oppression and ways to intervene at system levels to prevent problems. Medical legal partnership (MLP) clinics have done so by employing concepts from social work and health science programs on structural competency. This article examines MLP and related curriculum to meet the ABA mandate.

Assessment of Sanctions against Russian Science by Foreign Researchers: Diversity of Approaches

Introduction. The introduction of restrictive measures (sanctions) against Russian science, aimed at preventing Russia's participation in international scientific collaboration in both bilateral and multilateral formats, has become a prominent feature of modern international relations. For the first time in many years, international collaboration in various scientific fields has encountered significant political and economic constraints, leading to Russia's withdrawal from joint programs and projects with key partner states in the scientific sphere. The study's significance lies in the necessity to develop a theory related to this type of sanctions, given their unprecedented nature. A comprehensive understanding is needed, reflecting foreign scientists' views on the sanctions: whether they deem them justified and if they are willing to continue cooperation in any capacity. This will become a basis for modifying Russia's science and technology policy in the context of international cooperation. The subject of the study is the range of approaches foreign experts have toward the structure, scale, and implications of sanctions against Russian science. Object: Conceptual systematization of foreign researchers' perceptions regarding the phenomenon of anti-Russian sanctions in the scientific field. Tasks: identify the reasons behind the increased interest of foreign researchers and experts in the issue of international sanctions against Russian science; assess the level of consensus within the global scientific community regarding the need for such sanctions; identify the cause of bias among supporters of anti-Russian sanctions in the scientific field; determine the extent of foreign analysts' interest in the impact of sanctions on both Russian and global science; identify the pragmatic aspects of the current model of sanctions against Russian science. Degree of topic development: This type of research has not yet been conducted in either Russian or foreign science. Methods. General scientific methods (analysis and synthesis, abstraction and generalization, comparison), systemic and structural-functional approaches were used along with specialized scientific methods, such as discourse analysis, meta-analysis, and stochastic factor analysis technique. Result and Discussion. The article identifies three groups of approaches, along with their underlying arguments concerning the category of sanctions under consideration: radical, moderate, and liberal. The author demonstrates the predominance of a moderate approach, which aligns with official decisions to sever institutional scientific cooperation with Russia while permitting personal contacts. Conclusion. The international research community, aiming to gain a comprehensive understanding of the impact of sanctions against Russian science on Russia and the rest of the world, pursues two objectives. The first objective is to identify strategies for enhancing the effectiveness of sanctions and justifying their necessity while mitigating their negative impact on global science. The second objective is to find effective strategies for implementing scientific sanctions, which could potentially be applied against other countries in the future. Considering the positions and approaches of foreign researchers toward the phenomenon of anti-Russian scientific sanctions is crucial not only for developing the theory of scientific sanctions but also for shaping the concept of Russian science and technology policy under new circumstances. The novelty of this study lies in its first-ever attempt in both Russian and foreign scientific literature to conceptualize the range of foreign expert discourse, which is characterized by diverse views and approaches toward sanctions applied to Russian science. The article contributes to the relevant field of research, specifically the theory of international sanctions in the scientific sphere, by arguing for the need to complement the analysis of the objective content of sanctions with an analysis of their subjective evaluation by the expert community.

Evaluation of the presence of double-organ bias in published randomised controlled dacryocystorhinostomy studies.

The aim of this study is to show whether journals pay attention to double-organ bias in studies on dacryocystorhinostomy (DCR) treatment. Most statistical tests are based on the assumption that each data entry is independent of the other. However, we wanted to understand whether the eye, which is a double organ in the human body, is a cause of bias and whether sufficient attention has been paid to it in published articles. A PubMed search was conducted between November 1997 and November 2021 in the field of ophthalmology and otolaryngology with the terms 'dacryocystorhinostomy' and the limitations of 'people' and 'randomly controlled studies'. Publications containing DCR were searched. The publications published in the 15 journals with the highest impact were selected. The scope of the journal, the scope of the authors, the relevance of the articles, the presence of randomisation, the type of study (drug, endoscopic surgery, laser surgery and external surgery) and the statistical approach to bilateral organ bias were evaluated by two independent observers. A total of 83 publications were found by Pubmed search. Fifty-six publications were included in the study. Double-organ bias was present in 41% (n = 23) of the cases. Of the 23 biased publications, 46.4% were found to be in the Quartile 1 group. Quartile 4 journals showed bias in only four journals. Double-organ bias can also be seen in journals with high impact factors and published in recent years. Journals may need to pay more attention to double-organ bias in the publication acceptance process.

Healthcare bias and health inequalities towards displaced Syrians in Lebanon: a qualitative study.

According to Lebanese official data, Lebanon hosts over 1.5 million displaced Syrians (DS). Research shows that migrants encounter barriers when accessing healthcare. The social determinants of health (SDOH) related to migration are an additional challenge for DS in Lebanon, though bias plays a significant factor in exacerbating health inequalities. This study aims to identify DS perception of healthcare biases in the Lebanese healthcare system, and its consequences on DS' accessing and receiving quality healthcare in Lebanon. A qualitative analysis using in-depth, semi-structured interviews was utilized. 28 semi-structured interviews were conducted with doctors (n = 12) and nurses (n = 16) in 2021. Six group interviews were conducted with DS (n = 22) in Lebanese healthcare facilities. The recruitment of participants relied on reasoned and targeted sampling. Thematic analysis was performed to identify common themes in participants' experiences with DS accessing Lebanese healthcare. The findings indicated that there were barriers to accessing healthcare related to the SDOH, such as transportation and financial resources. The results also suggested that DS perceived health biases, including discriminatory behavior from Lebanese healthcare providers, stereotypes and racism leading to health inequalities. Based on the perceptions and experiences reported by participants, the underlying causes of biases are due to the fragility of the Lebanese healthcare system when facing a humanitarian crisis as well as a collapsing infrastructure torn by past wars and the current socio-political and financial crises in the country. Global initiatives are required to provide the necessary resources needed for offering equitable health services. Such initiatives involve addressing biases, health inequities, discrimination, and the lack of a Lebanese infrastructure system for the provision of healthcare. Addressing health inequalities remains a major health objective in achieving health equity on the micro level (cultural awareness and competencies) and macro level (equitable distribution of resources, implementation of a universal health coverage) in order to guarantee quality healthcare services to DS.

Fairness of artificial intelligence in healthcare: review and recommendations.

In this review, we address the issue of fairness in the clinical integration of artificial intelligence (AI) in the medical field. As the clinical adoption of deep learning algorithms, a subfield of AI, progresses, concerns have arisen regarding the impact of AI biases and discrimination on patient health. This review aims to provide a comprehensive overview of concerns associated with AI fairness; discuss strategies to mitigate AI biases; and emphasize the need for cooperation among physicians, AI researchers, AI developers, policymakers, and patients to ensure equitable AI integration. First, we define and introduce the concept of fairness in AI applications in healthcare and radiology, emphasizing the benefits and challenges of incorporating AI into clinical practice. Next, we delve into concerns regarding fairness in healthcare, addressing the various causes of biases in AI and potential concerns such as misdiagnosis, unequal access to treatment, and ethical considerations. We then outline strategies for addressing fairness, such as the importance of diverse and representative data and algorithm audits. Additionally, we discuss ethical and legal considerations such as data privacy, responsibility, accountability, transparency, and explainability in AI. Finally, we present the Fairness of Artificial Intelligence Recommendations in healthcare (FAIR) statement to offer best practices. Through these efforts, we aim to provide a foundation for discussing the responsible and equitable implementation and deployment of AI in healthcare.

Gender Bias in Artificial Intelligence

This study presents a Systematic Literature Review (SLR) of Gender Bias in Artificial Intelligence (AI). The research was conducted using two techniques: a domain-based approach to SLR process providing a bibliometric sample description and in-depth examination of the thematic categories arising from inductive categorization, extracted from reading and interpretation of the final 35 sample articles analyzed. In answering three key research questions on the types, causes, and overcoming (mitigating) strategies of gender bias in artificial intelligence, three thematic treemaps were constructed, enabling systematic overview as an essential contribution to the literature. The main types of gender bias found in AI are categorized as societal, technical, and individual. Societal and socio-technical aspects stand out as the leading causes of bias, while debiasing, dataset design and gender sensitivity were the most frequent among the main strategies for overcoming bias. The study also proposes theoretical, practical and managerial capacity building and policy implications that aim to influence broad socio-technical challenges and refer to changes necessary, aiming to create bias-free artificial intelligence.

Tropical Indian Ocean Mixed Layer Bias in CMIP6 CGCMs Primarily Attributed to the AGCM Surface Wind Bias

Abstract The relatively weak sea surface temperature bias in the tropical Indian Ocean (TIO) simulated in the coupled general circulation model (CGCM) from the recently released CMIP6 has been found to be important in model simulations of regional and global climate. However, the cause of the bias is debated because the bias is strongly model dependent and shows marked seasonality. In this study, we separate the bias in CGCMs into bias arising from oceanic GCMs (OGCMs) and bias that is independent of OGCMs using a set of CMIP6 and OMIP6 models. We found that OGCMs contribute little to mixed layer bias in the CGCMs. The OGCM-independent bias exhibits a large-scale cold mixed layer bias in the TIO throughout the year, with an unexpectedly high degree of model consistency. By conducting a set of OGCM experiments, we show that the OGCM-independent mixed layer bias is caused mainly by surface wind bias in the utilized CGCMs. About 89% of surface wind bias in the CGCMs is due to the inability of atmospheric GCMs (AGCMs), whereas atmosphere–ocean coupling in the CGCMs has only a minor influence on surface wind bias. The bias in surface wind is also found to be the cause of subsurface temperature bias besides the ocean dynamics such as vertical mixing and vertical shear in currents. Our results indicate that correcting TIO mixed layer bias in CGCMs requires improvement in the capability of AGCM in simulating the climatological surface winds. Significance Statement We aimed to discover the cause of temperature bias in the Indian Ocean in CMIP6 models. The bias was separated into oceanic model and ocean-model-independent bias to correspond exactly to bias caused by the oceanic model and by the atmospheric model and coupling, respectively. Oceanic bias contributes little to bias in CMIP6, but ocean-model-independent bias explains the CMIP6 bias throughout the year. We ran oceanic model experiments to show that surface wind bias causes ocean-model-independent temperature bias in the entire TIO and subsurface temperature bias in some areas of the Indian Ocean. We further found that 89% of surface wind bias originates from the atmospheric model. The results improve our understanding of the cause of the bias in the Indian Ocean and show that our method of bias separation is effective for attributing the source of bias to different proposed mechanisms.

Accuracy of Artificial Intelligence-Based Automated Quantitative Coronary Angiography Compared to Intravascular Ultrasound: Retrospective Cohort Study.

An accurate quantitative analysis of coronary artery stenotic lesions is essential to make optimal clinical decisions. Recent advances in computer vision and machine learning technology have enabled the automated analysis of coronary angiography. The aim of this paper is to validate the performance of artificial intelligence-based quantitative coronary angiography (AI-QCA) in comparison with that of intravascular ultrasound (IVUS). This retrospective study included patients who underwent IVUS-guided coronary intervention at a single tertiary center in Korea. Proximal and distal reference areas, minimal luminal area, percent plaque burden, and lesion length were measured by AI-QCA and human experts using IVUS. First, fully automated QCA analysis was compared with IVUS analysis. Next, we adjusted the proximal and distal margins of AI-QCA to avoid geographic mismatch. Scatter plots, Pearson correlation coefficients, and Bland-Altman were used to analyze the data. A total of 54 significant lesions were analyzed in 47 patients. The proximal and distal reference areas, as well as the minimal luminal area, showed moderate to strong correlation between the 2 modalities (correlation coefficients of 0.57, 0.80, and 0.52, respectively; P<.001). The correlation was weaker for percent area stenosis and lesion length, although statistically significant (correlation coefficients of 0.29 and 0.33, respectively). AI-QCA tended to measure reference vessel areas smaller and lesion lengths shorter than IVUS did. Systemic proportional bias was not observed in Bland-Altman plots. The biggest cause of bias originated from the geographic mismatch of AI-QCA with IVUS. Discrepancies in the proximal or distal lesion margins were observed between the 2 modalities, which were more frequent at the distal margins. After the adjustment of proximal or distal margins, there was a stronger correlation of proximal and distal reference areas between AI-QCA and IVUS (correlation coefficients of 0.70 and 0.83, respectively). AI-QCA showed a moderate to strong correlation compared with IVUS in analyzing coronary lesions with significant stenosis. The main discrepancy was in the perception of the distal margins by AI-QCA, and the correction of margins improved the correlation coefficients. We believe that this novel tool could provide confidence to treating physicians and help in making optimal clinical decisions.

Subsurface Thermohaline Biases in the Southern Tropical Pacific and the Roles of Wind Stress and Precipitation

Abstract Thermohaline structure and time evolution in the subsurface ocean play a critical role in climate variability and predictability. They are still poorly represented in ocean and climate models. Here, the characteristics of subsurface thermohaline biases in the southern tropical Pacific and their causes are investigated through CMIP-based analyses and model-based experiments. There exists a pronounced subsurface cold bias at 200-m depth over the southern tropical Pacific in CMIP6 simulations with an ensemble mean of about −4°C and an extreme close to −10°C. This cold bias is accompanied by a fresh subsurface bias of about −0.9 psu in the ensemble mean (−1.9 psu minimum). Similar subsurface thermohaline biases also exist in CMIP5 outputs, indicating that reduction of these biases remains a long-standing challenge for model developments. To understand the causes of these biases, attribution analyses and POP2-based sensitivity experiments are performed. It is found that the subsurface thermohaline biases are attributed to the model deficiencies in simulating wind stress curl and precipitation in the southern tropical Pacific. By conducting CESM2-based coupled experiments, a warm SST bias in the southeastern tropical Pacific is found to be responsible for the poor simulations in wind stress curl and precipitation. The consequences of these biases are also analyzed. The subsurface thermohaline biases cause the density field to increase substantially along 10°S, flattening the zonal isopycnal surface and reducing equatorward interior transport. In addition, the anomalously cold and fresh subsurface signals in the southern tropical Pacific are seen to propagate to the equator, leading to an overall spurious cooling in the equatorial subsurface. Significance Statement Subsurface biases severely degrade the credibility of climate models in their predictions and projections; hence, it is important to understand the causes of these subsurface biases. Our study analyzes the characteristics of subsurface thermohaline biases in the southern tropical Pacific and investigates their causes. A pronounced subsurface cold bias is found over the southern tropical Pacific, accompanied by an obvious subsurface fresh bias. By performing attribution analyses and numerical experiments, it is found that the subsurface thermohaline biases are attributed to the model deficiencies in simulating wind stress and precipitation, which are further attributed to the warm SST bias in the southeastern tropical Pacific. These results provide a guide for improving climate model performances.

FairMask: Better Fairness via Model-Based Rebalancing of Protected Attributes

<bold xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">Context</b> : Machine learning software can generate models that inappropriately discriminate against specific protected social groups (e.g., groups based on gender, ethnicity, etc.). Motivated by those results, software engineering researchers have proposed many methods for mitigating those discriminatory effects. While those methods are effective in mitigating bias, few of them can provide explanations on what is the root cause of bias. <bold xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">Objective</b> : We aim to better detect and mitigate algorithmic discrimination in machine learning software problems. <bold xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">Method</b> : Here we propose FairMask, a <italic xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">model-based</i> extrapolation method that is capable of both mitigating bias and explaining the cause. In our FairMask approach, protected attributes are represented by models learned from the other independent variables (and these models offer extrapolations over the space between existing examples). We then use the extrapolation models to relabel protected attributes later seen in testing data or deployment time. Our approach aims to offset the biased predictions of the classification model by rebalancing the distribution of protected attributes. <bold xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">Results</b> : The experiments of this paper show that, without compromising (original) model performance, FairMask can achieve significantly better group and individual fairness (as measured in different metrics) than benchmark methods. Moreover, compared to another instance-based rebalancing method, our model-based approach shows faster runtime and thus better scalability. <bold xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">Conclusion</b> : Algorithmic decision bias can be removed via extrapolation that corrects the misleading latent correlation between the protected attributes and other non-protected ones. As evidence for this, our proposed FairMask is not only performance-wise better (measured by fairness and performance metrics) than two state-of-the-art fairness algorithms. <bold xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">Reproduction Package</b> : In order to better support open science, all scripts and data used in this study are available online at <uri xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">https://github.com/anonymous12138/biasmitigation</uri> .

Using Risk of Bias 2 to assess results from randomised controlled trials: guidance from Cochrane

A systematic review identifies, appraises and synthesises all the empirical evidence from studies that meet prespecified eligibility criteria to answer a specific research question. As part of the appraisal, researchers...

Exploring Bias in Sclera Segmentation Models: A Group Evaluation Approach

Bias and fairness of biometric algorithms have been key topics of research in recent years, mainly due to the societal, legal and ethical implications of potentially unfair decisions made by automated decision-making models. A considerable amount of work has been done on this topic across different biometric modalities, aiming at better understanding the main sources of algorithmic bias or devising mitigation measures. In this work, we contribute to these efforts and present the first study investigating bias and fairness of sclera segmentation models. Although sclera segmentation techniques represent a key component of sclera-based biometric systems with a considerable impact on the overall recognition performance, the presence of different types of biases in sclera segmentation methods is still underexplored. To address this limitation, we describe the results of a group evaluation effort (involving seven research groups), organized to explore the performance of recent sclera segmentation models within a common experimental framework and study performance differences (and bias), originating from various demographic as well as environmental factors. Using five diverse datasets, we analyze seven independently developed sclera segmentation models in different experimental configurations. The results of our experiments suggest that there are significant differences in the overall segmentation performance across the seven models and that among the considered factors, ethnicity appears to be the biggest cause of bias. Additionally, we observe that training with representative and balanced data does not necessarily lead to less biased results. Finally, we find that in general there appears to be a negative correlation between the amount of bias observed (due to eye color, ethnicity and acquisition device) and the overall segmentation performance, suggesting that advances in the field of semantic segmentation may also help with mitigating bias.