Algorithmic Factors Influencing Bias in Machine Learning

Blanzeisky, William; Cunningham, Pádraig

doi:10.1007/978-3-030-93736-2_41

Cited by 16 publications

(10 citation statements)

References 17 publications

Supporting

Mentioning

Contrasting

Order By: Relevance

“…Evaluations on bias in ML are necessarily limited because of the availability of relevant datasets. Our evaluation considers the synthetic dataset introduced in Feldman et al (2015), the synthetic exemplar dataset in Blanzeisky et al (2021), the reduced version of the Adult dataset (Kohavi, 1996) and the ProPublica Recidivism dataset (Dressel & Farid, 2018). These datasets have been extensively studied in fairness research because there is clear evidence of negative legacy.…”

Section: Methodsmentioning

confidence: 99%

“…On the other hand, underestimation occurs when the algorithm focuses on strong signals in the data thereby missing more subtle phenomena (Cunningham & Delany, 2020). Recent work shows that underestimation occurs when an algorithm underfits the training data due to a combination of limitations in training data and model capacity issues (Kamishima et al , 2012; Cunningham & Delany, 2020; Blanzeisky & Cunningham, 2021). It has also been shown that irreducible error, regularization and feature and class imbalance can contribute to this underestimation (Blanzeisky & Cunningham, 2021).…”

Section: Bias In Machine Learningmentioning

confidence: 99%

“…Thus, algorithms that are trained on a biased representation of the actual population could disproportionately disadvantage a specific group or groups. Although most examples of algorithmic bias in ML occur due to the training data, recent work shows that the algorithm itself can introduce bias or amplify existing bias (Kamishima et al , 2012; Baeza-Yates, 2018; Cunningham & Delany, 2020; Hooker et al , 2020; Blanzeisky & Cunningham, 2021). When the bias occurs due to the data, it is sometimes euphemistically called negative legacy; it is called underestimation when it is due to the algorithm (Kamishima et al , 2012; Cunningham & Delany, 2020).…”

Section: Introductionmentioning

confidence: 99%

See 2 more Smart Citations

Using Pareto simulated annealing to address algorithmic bias in machine learning

Blanzeisky¹,

Cunningham²

2022

The Knowledge Engineering Review

Self Cite

View full text Add to dashboard Cite

Algorithmic bias arises in machine learning when models that may have reasonable overall accuracy are biased in favor of ‘good’ outcomes for one side of a sensitive category, for example gender or race. The bias will manifest as an underestimation of good outcomes for the under-represented minority. In a sense, we should not be surprised that a model might be biased when it has not been ‘asked’ not to be; reasonable accuracy can be achieved by ignoring the under-represented minority. A common strategy to address this issue is to include fairness as a component in the learning objective. In this paper, we consider including fairness as an additional criterion in model training and propose a multi-objective optimization strategy using Pareto Simulated Annealing that optimizes for both accuracy and underestimation bias. Our experiments show that this strategy can identify families of models with members representing different accuracy/fairness tradeoffs. We demonstrate the effectiveness of this strategy on two synthetic and two real-world datasets.

show abstract

Section: Methodsmentioning

confidence: 99%

Section: Bias In Machine Learningmentioning

confidence: 99%

Section: Introductionmentioning

confidence: 99%

See 1 more Smart Citation

Using Pareto simulated annealing to address algorithmic bias in machine learning

Blanzeisky¹,

Cunningham²

2022

The Knowledge Engineering Review

Self Cite

View full text Add to dashboard Cite

show abstract

“…Algorithmic bias in ML-based models stems from abnormal datasets, weak models, poor algorithm designs, or historical human biases [27]. Algorithmic bias also can happen due to the problem of under-fitting in the training phase, which can be caused by a mixture of limitations in the training data and model capacity issues [28]. The factors affecting this mechanism are irreducible error (Bayes error), regularization mechanisms, class imbalance, and under-represented categories [28].…”

Section: Bias In Machine Learningmentioning

confidence: 99%

A Comparative Analysis of Bias Amplification in Graph Neural Network Approaches for Recommender Systems

2022

View full text Add to dashboard Cite

Recommender Systems (RSs) are used to provide users with personalized item recommendations and help them overcome the problem of information overload. Currently, recommendation methods based on deep learning are gaining ground over traditional methods such as matrix factorization due to their ability to represent the complex relationships between users and items and to incorporate additional information. The fact that these data have a graph structure and the greater capability of Graph Neural Networks (GNNs) to learn from these structures has led to their successful incorporation into recommender systems. However, the bias amplification issue needs to be investigated while using these algorithms. Bias results in unfair decisions, which can negatively affect the company’s reputation and financial status due to societal disappointment and environmental harm. In this paper, we aim to comprehensively study this problem through a literature review and an analysis of the behavior against biases of different GNN-based algorithms compared to state-of-the-art methods. We also intend to explore appropriate solutions to tackle this issue with the least possible impact on the model’s performance.

show abstract

“…gender) relative to its multiple values (e.g. male versus female), then model predictions will also favour the most represented attribute class while the underestimation for the minority class will be accentuated [10].…”

Section: The Need For Unbiased Modelsmentioning

confidence: 99%

MarIA and BETO are sexist: evaluating gender bias in large language models for Spanish

Garrido-Muñoz

Martínez-Santiago

Montejo-Ráez

2022

Preprint

View full text Add to dashboard Cite

The study of bias in language models is a growing area of work, however, both research and resources are focused on English. In this paper, we make a first approach focusing on gender bias in some freely available Spanish language models trained using popular deep neural networks, like BERT or RoBERTa. Some of these models are known for achieving state-of-the-art results on downstream tasks. These promising results have promoted such models' integration in many real-world applications and production environments, which could be detrimental to people affected for those systems. This work proposes an evaluation framework to identify gender bias in masked language models, with explainability in mind to ease the interpretation of the evaluation results. We have evaluated 20 different models for Spanish, including some of the most popular pretrained ones in the research community. Our findings state that varying levels of gender bias are present across these models. This approach compares the adjectives proposed by the model for a set of templates. We classify the given adjectives into understandable categories and compute two new metrics from model predictions, one based on the internal state (probability) and the other one on the external state (rank). Those metrics are used to reveal biased models according to the given categories and quantify the degree of bias of the models under study.

show abstract

Algorithmic Factors Influencing Bias in Machine Learning

Cited by 16 publications

References 17 publications

Using Pareto simulated annealing to address algorithmic bias in machine learning

Using Pareto simulated annealing to address algorithmic bias in machine learning

A Comparative Analysis of Bias Amplification in Graph Neural Network Approaches for Recommender Systems

MarIA and BETO are sexist: evaluating gender bias in large language models for Spanish

Contact Info

Product

Resources

About