8,882 Hits in 4.0 sec

Balanced Datasets Are Not Enough: Estimating and Mitigating Gender Bias in Deep Image Representations [article]

Tianlu Wang, Jieyu Zhao, Mark Yatskar, Kai-Wei Chang, Vicente Ordonez
2019 arXiv   pre-print
To mitigate this, we adopt an adversarial approach to remove unwanted features corresponding to protected variables from intermediate representations in a deep neural network -- and provide a detailed  ...  We show that trained models significantly amplify the association of target labels with gender beyond what one would expect from biased datasets.  ...  We also acknowledge fruitful discussions with members of the Human-Machine Intelligence group through the Institute for the Humanities and Global Cultures at the University of Virginia.  ... 
arXiv:1811.08489v4 fatcat:f34iwjfusfeqdfdsbyumx7yir4

Jointly De-biasing Face Recognition and Demographic Attribute Estimation [article]

Sixue Gong, Xiaoming Liu, Anil K. Jain
2020 arXiv   pre-print
Adversarial learning is adopted to minimize correlation among feature factors so as to abate bias influence from other factors.  ...  We present a novel de-biasing adversarial network (DebFace) that learns to extract disentangled feature representations for both unbiased face recognition and demographics estimation.  ...  Given this assumption, one common way to remove demographic information from face representations is to perform feature disentanglement via adversarial learning (Fig. 1b) .  ... 
arXiv:1911.08080v4 fatcat:2cavhrnfezggjh6jffhpxzhfoy

Linear Adversarial Concept Erasure [article]

Shauli Ravfogel, Michael Twiton, Yoav Goldberg, Ryan Cotterell
2022 arXiv   pre-print
When evaluated in the context of binary gender removal, the method recovers a low-dimensional subspace whose removal mitigates bias by intrinsic and extrinsic evaluation.  ...  We formulate the problem of identifying and erasing a linear subspace that corresponds to a given concept, in order to prevent linear predictors from recovering the concept.  ...  gender at all. 5 Besides the Figure 4 : Application of R-LACE on the raw pixels of image data, from top to bottom we present the original images and the same images after a rank-1 projection, for the  ... 
arXiv:2201.12091v1 fatcat:qfcltkbdlrayhhk6bj6y7ix55e

SensitiveNets: Learning Agnostic Representations with Application to Face Images [article]

Aythami Morales, Julian Fierrez, Ruben Vera-Rodriguez, Ruben Tolosana
2020 arXiv   pre-print
Our method is based on an adversarial regularizer that introduces a sensitive information removal function in the learning objective.  ...  Instead of existing approaches aimed directly at fairness improvement, the proposed feature representation enforces the privacy of selected attributes.  ...  removing the sensitive information (gender and ethnicity in this case) from the embeddings.  ... 
arXiv:1902.00334v3 fatcat:d57brut44vg2zoeppqwx24b5cq

Fairness in Deep Learning: A Computational Perspective [article]

Mengnan Du, Fan Yang, Na Zou, Xia Hu
2020 arXiv   pre-print
We provide a review covering recent progresses to tackle algorithmic fairness problems of deep learning from the computational perspective.  ...  We also discuss fairness mitigation approaches categorized according to three stages of deep learning life-cycle, aiming to push forward the area of fairness in deep learning and build genuinely fair and  ...  Thus prediction outcome discrimination could be detected and removed from the deep representation perspective.  ... 
arXiv:1908.08843v2 fatcat:kaaevm64fbctpjfdycv5uz3dhi

Privacy-Preserving Image Template Sharing Using Contrastive Learning

Shideh Rezaeifar, Slava Voloshynovskiy, Meisam Asgari Asgari Jirhandeh, Vitality Kinakh
2022 Entropy  
For the attribute inference attack, we aim to provide a representation of data that is independent of the sensitive attribute.  ...  Furthermore, an obfuscator module is trained in an adversarial manner to preserve the privacy of sensitive attributes while maintaining the classification performance on the target attribute.  ...  Deep and private feature sharing: With the recent advancements of deep models, a new line of work has been introduced to share deep private and obfuscated feature representations of images. Osia et al  ... 
doi:10.3390/e24050643 fatcat:z6svnfowzbe3zas4jftbpzy46m

Privacy-Preserving Image Acquisition Using Trainable Optical Kernel [article]

Yamin Sepehri, Pedram Pad, Pascal Frossard, L. Andrea Dunbar
2021 arXiv   pre-print
In this work, for the first time, we propose a trainable image acquisition method that removes the sensitive identity revealing information in the optical domain before it reaches the image sensor.  ...  We show that this method can reduce 65.1% of sensitive content when it is selected to be the gender and it only loses 7.3% of the desired content.  ...  After training, the visual representation inverter can be used to reconstruct the original images from the output of the optical convolution.  ... 
arXiv:2106.14577v1 fatcat:o5hzef4sejcovlptldclvpy4t4

Toward Privacy and Utility Preserving Image Representation [article]

Ahmadreza Mosallanezhad and Yasin N. Silva and Michelle V. Mancenido and Huan Liu
2020 arXiv   pre-print
Multiple methods have been proposed to protect an individual's privacy by perturbing the images to remove traces of identifiable information, such as gender or race.  ...  In this paper, we study the novel problem of creating privacy-preserving image representations with respect to a given utility task by proposing a principled framework called the Adversarial Image Anonymizer  ...  For example, a well-trained gender classifier could predict the gender of a person using the corresponding image representation.  ... 
arXiv:2009.14376v2 fatcat:euznouscvvarjcubasn3ma5qz4

Training privacy-preserving video analytics pipelines by suppressing features that reveal information about private attributes [article]

Chau Yi Li, Andrea Cavallaro
2022 arXiv   pre-print
attributes (e.g. age or gender).  ...  Deep neural networks are increasingly deployed for scene analytics, including to evaluate the attention and reaction of people exposed to out-of-home advertisements.  ...  The ARM network aims to remove the distortion from the edges of the image on the features, referred to as albino erosion.  ... 
arXiv:2203.02635v1 fatcat:vwofqa4b3jetvjao6g7chxpwlq

Deep Identity-aware Transfer of Facial Attributes [article]

Mu Li and Wangmeng Zuo and David Zhang
2018 arXiv   pre-print
This paper presents a Deep convolutional network model for Identity-Aware Transfer (DIAT) of facial attributes.  ...  Our DIAT can provide a unified solution for several representative facial attribute transfer tasks, e.g., expression transfer, accessory removal, age progression, and gender transfer, and can be extended  ...  transfer, accessory removal, age progression, and gender transfer.  ... 
arXiv:1610.05586v2 fatcat:l4khpvjrnbgh3k6gy7csfyzfse

Towards Reducing Bias in Gender Classification [article]

Komal K. Teru, Aishik Chakraborty
2019 arXiv   pre-print
This work aims at addressing the racial bias present in many modern gender recognition systems. We learn race invariant representations of human faces with an adversarially trained autoencoder model.  ...  We show that such representations help us achieve less biased performance in gender classification.  ...  In particular, we remove race information from the representation of each image and try to do gender classification using this disentangled representation.  ... 
arXiv:1911.08556v1 fatcat:7nptsrjej5blrdgiaaopfhza24

Representation Learning with Statistical Independence to Mitigate Bias [article]

Ehsan Adeli, Qingyu Zhao, Adolf Pfefferbaum, Edith V. Sullivan, Li Fei-Fei, Juan Carlos Niebles, Kilian M. Pohl
2020 arXiv   pre-print
Such challenges range from spurious associations between variables in medical studies to the bias of race in gender or face recognition systems.  ...  We apply our method to synthetic data, medical images (containing task bias), and a dataset for gender classification (containing dataset bias).  ...  ., gender prediction from face images.  ... 
arXiv:1910.03676v4 fatcat:yf6amgpfivgarmfunohtx3l2ry

Don't Judge Me by My Face : An Indirect Adversarial Approach to Remove Sensitive Information From Multimodal Neural Representation in Asynchronous Job Video Interviews [article]

Léo Hemamou, Arthur Guillon, Jean-Claude Martin, Chloé Clavel
2021 arXiv   pre-print
Recently, adversarial methods have been proved to effectively remove sensitive information from the latent representation of neural networks.  ...  In this article, we propose a new adversarial approach to remove sensitive information from the latent representation of neural networks without the need to collect any sensitive variable.  ...  PROPOSED CONTRIBUTION: AN INDIRECT ADVERSARIAL APPROACH We present a new adversarial approach, available in two forms, to remove personal information of AVI candidates from the latent layer of a deep network  ... 
arXiv:2110.09424v1 fatcat:iknfspsudbdvjdqfwvpvvcg5t4

TIPRDC: Task-Independent Privacy-Respecting Data Crowdsourcing Framework for Deep Learning with Anonymized Intermediate Representations [article]

Ang Li, Yixiao Duan, Huanrui Yang, Yiran Chen, Jianlei Yang
2020 arXiv   pre-print
The success of deep learning partially benefits from the availability of various large-scale datasets.  ...  The emerging privacy concerns from users on data sharing hinder the generation or use of crowdsourcing datasets and lead to hunger of training data for new deep learning applications.  ...  ACKNOWLEDGMENTS This work was supported in part by NSF-1822085 and NSF IUCRC for ASIC membership from Ergomotion.  ... 
arXiv:2005.11480v6 fatcat:jgppdb5fanby5ocdjrjkvpeoeq

DISCO: Dynamic and Invariant Sensitive Channel Obfuscation for deep neural networks [article]

Abhishek Singh, Ayush Chopra, Vivek Sharma, Ethan Garza, Emily Zhang, Praneeth Vepakomma, Ramesh Raskar
2021 arXiv   pre-print
Recent deep learning models have shown remarkable performance in image classification.  ...  Finally, we also release an evaluation benchmark dataset of 1 million sensitive representations to encourage rigorous exploration of novel attack schemes.  ...  A key idea of DISCO is to disentangle representation learning from privacy via the learned pruning filter.  ... 
arXiv:2012.11025v2 fatcat:dqg6gkx7mvh7dka3sqdcufcwd4
« Previous Showing results 1 — 15 out of 8,882 results