Investigating Bias in Facial Analysis Systems: A Systematic Review

Khalil, Ashraf; Ahmed, Soha; Khattak, Asad Masood; Al-Qirim, Nabeel

doi:10.1109/access.2020.3006051

Cited by 49 publications

(41 citation statements)

References 41 publications

(78 reference statements)

Supporting

Mentioning

Contrasting

Order By: Relevance

“…Over a half of a decade later, there has been at least one significant contribution in conventional FR each year ever since (i.e., 2015 [14], 2016 [15], 2017 [8], 2018 [7], 2019 [16], and even 2020, the year of facial masks [17]). Scholars provide details on deep learning advances in FR technology as a part of recent surveys [18], [19], with another focused on bias specific to FR [20].…”

Section: Bias In Frmentioning

confidence: 99%

Balancing Biases and Preserving Privacy on Balanced Faces in the Wild

Robinson¹,

Qin²,

Henon³

et al. 2021

Preprint

View full text Add to dashboard Cite

There are demographic biases in current models used for facial recognition (FR). Our Balanced Faces In the Wild (BFW) dataset serves as a proxy to measure bias across ethnicity and gender subgroups, allowing one to characterize FR performances per subgroup. We show performances are non-optimal when a single score threshold is used to determine whether sample pairs are genuine or imposter. Across subgroups, performance ratings vary from the reported across the entire dataset. Thus, claims of specific error rates only hold true for populations matching that of the validation data. We mitigate the imbalanced performances using a novel domain adaptation learning scheme on the facial features extracted using state-of-the-art. Not only does this technique balance performance, but it also boosts the overall performance. A benefit of the proposed is to preserve identity information in facial features while removing demographic knowledge in the lower dimensional features. The removal of demographic knowledge prevents future potential biases from being injected into decision-making. This removal satisfies privacy concerns. We explore why this works qualitatively; we also show quantitatively that subgroup classifiers can no longer learn from the features mapped by the proposed.

show abstract

Section: Bias In Frmentioning

confidence: 99%

Balancing Biases and Preserving Privacy on Balanced Faces in the Wild

Robinson¹,

Qin²,

Henon³

et al. 2021

Preprint

View full text Add to dashboard Cite

show abstract

“…Many biometrics exist to provide authentication for users while in a public setting [1], such as personal identification numbers, passwords, cards, keys, and tokens [2]. However, those methods can become compromised, lost, duplicated, stolen, or challenging to recall [2].…”

Section: Introductionmentioning

confidence: 99%

“…Many users want authentication in a public setting and most likely, using a mobile device leads to unconstructed environments [6] and non-controlled changes [11]. These changes lead to limitations on nonlinear variations [11], making data acquisition difficult [1]. This problem has persisted for over fifty years [6] and often contributes to differing results that cannot be replicated [10].…”

Section: Introductionmentioning

confidence: 99%

“…Further complications include the approach taken because different techniques can yield different results [10]. Some influences that contribute to these problems are the position, illumination, and expression [1][2][3][4]8,12,13]. Other influences include pose angles [1,8], camera distance, head size, rotation, angle, and direction [3,5,6].…”

Section: Introductionmentioning

confidence: 99%

“…Some influences that contribute to these problems are the position, illumination, and expression [1][2][3][4]8,12,13]. Other influences include pose angles [1,8], camera distance, head size, rotation, angle, and direction [3,5,6]. The introduction of aging, scaling, accessories, occlusion [1,13], and hair [5] makes capturing these varying scales more difficult [14].…”

Section: Introductionmentioning

confidence: 99%

See 2 more Smart Citations

Evaluating Impact of Race in Facial Recognition across Machine Learning and Deep Learning Algorithms

Coe

Atay

2021

Computers

View full text Add to dashboard Cite

The research aims to evaluate the impact of race in facial recognition across two types of algorithms. We give a general insight into facial recognition and discuss four problems related to facial recognition. We review our system design, development, and architectures and give an in-depth evaluation plan for each type of algorithm, dataset, and a look into the software and its architecture. We thoroughly explain the results and findings of our experimentation and provide analysis for the machine learning algorithms and deep learning algorithms. Concluding the investigation, we compare the results of two kinds of algorithms and compare their accuracy, metrics, miss rates, and performances to observe which algorithms mitigate racial bias the most. We evaluate racial bias across five machine learning algorithms and three deep learning algorithms using racially imbalanced and balanced datasets. We evaluate and compare the accuracy and miss rates between all tested algorithms and report that SVC is the superior machine learning algorithm and VGG16 is the best deep learning algorithm based on our experimental study. Our findings conclude the algorithm that mitigates the bias the most is VGG16, and all our deep learning algorithms outperformed their machine learning counterparts.

show abstract

Professionalism in artificial intelligence: The link between technology and ethics

Klarin,

Ali Abadi,

Sharmelly

2024

Syst Res Behav Sci

View full text Add to dashboard Cite

Ethical conduct of artificial intelligence (AI) is undoubtedly becoming an ever more pressing issue considering the inevitable integration of these technologies into our lives. The literature so far discussed the responsibility domains of AI; this study asks the question of how to instil ethicality into AI technologies. Through a three‐step review of the AI ethics literature, we find that (i) the literature is weak in identifying solutions in ensuring ethical conduct of AI, (ii) the role of professional conduct is underexplored, and (iii) based on the values extracted from studies about AI ethical breaches, we thus propose a conceptual framework that offers professionalism as a solution in ensuring ethical AI. The framework stipulates fairness, nonmaleficence, responsibility, freedom, and trust as values necessary for developers and operators, as well as transparency, privacy, fairness, trust, solidarity, and sustainability as organizational values to ensure sustainability in ethical development and operation of AI.

show abstract

Investigating Bias in Facial Analysis Systems: A Systematic Review

Cited by 49 publications

References 41 publications

Balancing Biases and Preserving Privacy on Balanced Faces in the Wild

Balancing Biases and Preserving Privacy on Balanced Faces in the Wild

Evaluating Impact of Race in Facial Recognition across Machine Learning and Deep Learning Algorithms

Professionalism in artificial intelligence: The link between technology and ethics

Contact Info

Product

Resources

About