Fairness GAN: Generating datasets with fairness properties using a generative adversarial network

Sattigeri, Prasanna; Hoffman, Samuel C.; Chenthamarakshan, Vijil; Varshney, Kush R.

doi:10.1147/jrd.2019.2945519

Cited by 132 publications

(91 citation statements)

References 15 publications

Supporting

Mentioning

Contrasting

Order By: Relevance

“…Generating debiased data: Recent approaches have utilized generative models [30] and data transformations [11] to generate training data that is more 'fair' than the original dataset. For example, Sattigeri et al [30] used a generative adversarial network (GAN) to output a reconstructed dataset similar to the input but more fair with respect to certain attributes. Pre-processing data transformations that mitigate discrimination [11] have also been proposed, yet such methods are not learned adaptively during training nor do they provide realistic training examples.…”

Section: Related Workmentioning

confidence: 99%

Uncovering and Mitigating Algorithmic Bias through Learned Latent Structure

Soleimany

Schwarting

Bhatia

et al. 2019

Proceedings of the 2019 AAAI/ACM Conference on AI, Ethics, and Society

201

184

View full text Add to dashboard Cite

Recent research has highlighted the vulnerabilities of modern machine learning based systems to bias, especially towards segments of society that are under-represented in training data. In this work, we develop a novel, tunable algorithm for mitigating the hidden, and potentially unknown, biases within training data. Our algorithm fuses the original learning task with a variational autoencoder to learn the latent structure within the dataset and then adaptively uses the learned latent distributions to re-weight the importance of certain data points while training. While our method is generalizable across various data modalities and learning tasks, in this work we use our algorithm to address the issue of racial and gender bias in facial detection systems. We evaluate our algorithm on the Pilot Parliaments Benchmark (PPB), a dataset specifically designed to evaluate biases in computer vision systems, and demonstrate increased overall performance as well as decreased categorical bias with our debiasing approach. CCS CONCEPTS • Computing methodologies → Artificial intelligence; Machine learning; Neural networks.

show abstract

Section: Related Workmentioning

confidence: 99%

Uncovering and Mitigating Algorithmic Bias through Learned Latent Structure

Soleimany

Schwarting

Bhatia

et al. 2019

Proceedings of the 2019 AAAI/ACM Conference on AI, Ethics, and Society

201

184

View full text Add to dashboard Cite

show abstract

“…Nowadays, generative adversarial networks are largely investigated; the results showed that GANs could create highly realistic images from scratch [49], [50]. One of the major difficulties in learning facial beauty is that it must be carried out in an unsupervised manner because there are no fewer or more pairs of attractive images of an equal individual that may require supervised learning [51].…”

Section: Bfp and Generative Adversarial Network (Gans)mentioning

confidence: 99%

Facial Beauty Prediction and Analysis based on Deep Convolutional Neural Network: A Review

Saeed

Abdulazeez

2021

JSCDM

View full text Add to dashboard Cite

Facial attractiveness or facial beauty prediction (FBP) is a current study that has several potential usages. It is a key difficulty area in the computer vision domain because of the few public databases related to FBP and its experimental trials on the minor-scale database. Moreover, the evaluation of facial beauty is personalized in nature, with people having personalized favor of beauty. Deep learning techniques have displayed a significant ability in terms of analysis and feature representation. The previous studies focussed on scattered portions of facial beauty with fewer comparisons between diverse techniques. Thus, this article reviewed the recent research on computer prediction and analysis of face beauty based on deep convolution neural network DCNN. Furthermore, the provided possible lines of research and challenges in this article can help researchers in advancing the state – of- art in future work.

show abstract

“…These methods employ techniques from domain adaptation to learn a representation that minimizes classification loss while being invariant to the sensitive attribute. In the lat-ter category, Sattigeri et al (2018) extend AC-GAN (Odena et al 2017) to generate a fair dataset, while Quadrianto et al (2018) use an autoencoder to remove sensitive information from images. In this work, we introduce an image-to-image translation model to augment the training set, and thus, our framework is most closely related to Quadrianto et al (2018).…”

Section: Fairness-aware Learning and Face Analysismentioning

confidence: 99%

Mitigating Demographic Bias in Facial Datasets with Style-Based Multi-attribute Transfer

et al. 2021

View full text Add to dashboard Cite

Deep learning has catalysed progress in tasks such as face recognition and analysis, leading to a quick integration of technological solutions in multiple layers of our society. While such systems have proven to be accurate by standard evaluation metrics and benchmarks, a surge of work has recently exposed the demographic bias that such algorithms exhibit–highlighting that accuracy does not entail fairness. Clearly, deploying biased systems under real-world settings can have grave consequences for affected populations. Indeed, learning methods are prone to inheriting, or even amplifying the bias present in a training set, manifested by uneven representation across demographic groups. In facial datasets, this particularly relates to attributes such as skin tone, gender, and age. In this work, we address the problem of mitigating bias in facial datasets by data augmentation. We propose a multi-attribute framework that can successfully transfer complex, multi-scale facial patterns even if these belong to underrepresented groups in the training set. This is achieved by relaxing the rigid dependence on a single attribute label, and further introducing a tensor-based mixing structure that captures multiplicative interactions between attributes in a multilinear fashion. We evaluate our method with an extensive set of qualitative and quantitative experiments on several datasets, with rigorous comparisons to state-of-the-art methods. We find that the proposed framework can successfully mitigate dataset bias, as evinced by extensive evaluations on established diversity metrics, while significantly improving fairness metrics such as equality of opportunity.

show abstract

Fairness GAN: Generating datasets with fairness properties using a generative adversarial network

Cited by 132 publications

References 15 publications

Uncovering and Mitigating Algorithmic Bias through Learned Latent Structure

Uncovering and Mitigating Algorithmic Bias through Learned Latent Structure

Facial Beauty Prediction and Analysis based on Deep Convolutional Neural Network: A Review

Mitigating Demographic Bias in Facial Datasets with Style-Based Multi-attribute Transfer

Contact Info

Product

Resources

About