SenSeI: Sensitive Set Invariance for Enforcing Individual Fairness

Yurochkin, Mikhail; Sun, Yuekai

doi:10.48550/arxiv.2006.14168

Cited by 6 publications

(17 citation statements)

References 25 publications

Supporting

Mentioning

Contrasting

Order By: Relevance

“…In this section, we conduct experiments on various datasets to fully demonstrate the effectiveness of the proposed method. Experimental settings and datasets adopted by debiasing and fairness papers are different from each other, and we mainly follow three different settings as [25,41,1], [35] and [53,54]. We conduct experiments on Colored MNIST [25], IMDB face [44], CelebA [33], mPower [9], and Adult [2].…”

Section: Methodsmentioning

confidence: 99%

“…We conduct experiments on Colored MNIST [25], IMDB face [44], CelebA [33], mPower [9], and Adult [2]. For Colored MNIST, IMDB face, and mPower, we follow the debiasing setting adopted by [25,41,1], for CelebA, we follow the setting adopted by [35], and for Adult, we follow the fairness setting adopted by [53,54]. Among these datasets, Colored MNIST, IMDB Face, and CelebA are image datasets, mPower is a time series dataset, and Adult is a tabular dataset.…”

Section: Methodsmentioning

confidence: 99%

“…SenSR adopts a variant of individual fairness as a regularizer so that the learned model could satisfy the individual fairness [53]. Similarly, SenSeI achieves the individual fairness through a transport-based regularizer [54]. Zafar et al develop fairness methods based on the decision boundary fairness [55].…”

Section: Debiasing and Fairnessmentioning

confidence: 99%

“…We first preprocess the dataset following [53], and split the data into 80% training and 20% testing. We compare our method with advanced fairness methods including Project [53], CoCL [15], Adversarial Debiasing [56], SenSR [53], and SenSeI [54]. The results of compared methods are duplicated from [54], and are obtained with a two-layer MLP for the target task.…”

Section: Adultmentioning

confidence: 99%

“…We compare our method with advanced fairness methods including Project [53], CoCL [15], Adversarial Debiasing [56], SenSR [53], and SenSeI [54]. The results of compared methods are duplicated from [54], and are obtained with a two-layer MLP for the target task. However, recall that CSAD contains three different modules for target task including F , D y , and P y , which make the twolayer MLP not applicable for our method, we thus construct a three-layer MLP which has similar performance as shown in Table 5.…”

Section: Adultmentioning

confidence: 99%

See 4 more Smart Citations

Learning Bias-Invariant Representation by Cross-Sample Mutual Information Minimization

Zhu¹,

Zheng²,

Liao³

et al. 2021

Preprint

View full text Add to dashboard Cite

Deep learning algorithms mine knowledge from the training data and thus would likely inherit the dataset's bias information. As a result, the obtained model would generalize poorly and even mislead the decision process in real-life applications. We propose to remove the bias information misused by the target task with a crosssample adversarial debiasing (CSAD) method. CSAD explicitly extracts target and bias features disentangled from the latent representation generated by a feature extractor and then learns to discover and remove the correlation between the target and bias features. The correlation measurement plays a critical role in adversarial debiasing and is conducted by a cross-sample neural mutual information estimator. Moreover, we propose joint content and local structural representation learning to boost mutual information estimation for better performance. We conduct thorough experiments on publicly available datasets to validate the advantages of the proposed method over state-of-the-art approaches.

show abstract

Section: Methodsmentioning

confidence: 99%

Section: Methodsmentioning

confidence: 99%

Section: Debiasing and Fairnessmentioning

confidence: 99%

Section: Adultmentioning

confidence: 99%

Section: Adultmentioning

confidence: 99%

See 3 more Smart Citations

Learning Bias-Invariant Representation by Cross-Sample Mutual Information Minimization

Zhu¹,

Zheng²,

Liao³

et al. 2021

Preprint

View full text Add to dashboard Cite

show abstract

Removing Spurious Features can Hurt Accuracy and Affect Groups Disproportionately

Khani

Liang

2021

Proceedings of the 2021 ACM Conference on Fairness, Accountability, and Transparency

View full text Add to dashboard Cite

The presence of spurious features interferes with the goal of obtaining robust models that perform well across many groups within the population. A natural remedy is to remove spurious features from the model. However, in this work we show that removal of spurious features can decrease accuracy due to the inductive biases of overparameterized models. We completely characterize how the removal of spurious features affects accuracy across different groups (more generally, test distributions) in noiseless overparameterized linear regression. In addition, we show that removal of spurious feature can decrease the accuracy even in balanced dataset-each target co-occurs equally with each spurious feature; and it can inadvertently make the model more susceptible to other spurious features. Finally, we show that robust self-training can remove spurious features without affecting the overall accuracy. Experiments on the Toxic-Comment-Detectoin and CelebA datasets show that our results hold in non-linear models.

show abstract

SLIDE: a surrogate fairness constraint to ensure fairness consistency

Kunwoong¹,

Ohn²,

Kim³

et al. 2022

Preprint

View full text Add to dashboard Cite

As they have a vital effect on social decision makings, AI algorithms should be not only accurate and but also fair. Among various algorithms for fairness AI, learning a prediction model by minimizing the empirical risk (e.g., crossentropy) subject to a given fairness constraint has received much attention. To avoid computational difficulty, however, a given fairness constraint is replaced by a surrogate fairness constraint as the 0-1 loss is replaced by a convex surrogate loss for classification problems. In this paper, we investigate the validity of existing surrogate fairness constraints and propose a new surrogate fairness constraint called SLIDE, which is computationally feasible and asymptotically valid in the sense that the learned model satisfies the fairness constraint asymptotically and achieves a fast convergence rate. Numerical experiments confirm that the SLIDE works well for various benchmark datasets.

show abstract

SenSeI: Sensitive Set Invariance for Enforcing Individual Fairness

Cited by 6 publications

References 25 publications

Learning Bias-Invariant Representation by Cross-Sample Mutual Information Minimization

Learning Bias-Invariant Representation by Cross-Sample Mutual Information Minimization

Removing Spurious Features can Hurt Accuracy and Affect Groups Disproportionately

SLIDE: a surrogate fairness constraint to ensure fairness consistency

Contact Info

Product

Resources

About