{"about":{"site":"https://codewithpapers.app","non_affiliation":"Code with Papers and Syntology are not affiliated with, endorsed by, or sponsored by Papers with Code, Meta, or the pwc-archive mirror.","licence":"CC BY-SA 4.0","licence_url":"https://creativecommons.org/licenses/by-sa/4.0/legalcode","attribution":"https://codewithpapers.app/attribution","modified":"archive material modified by Syntology; see the attribution page"},"url":"/paper/neural-styling-for-interpretable-fair","title":"Discovering Fair Representations in the Data Domain","arxiv_id":"1810.06755","date":"2018-10-15","proceeding":"CVPR 2019 6","authors":["Novi Quadrianto","Viktoriia Sharmanska","Oliver Thomas"],"abstract":"Interpretability and fairness are critical in computer vision and machine\nlearning applications, in particular when dealing with human outcomes, e.g.\ninviting or not inviting for a job interview based on application materials\nthat may include photographs. One promising direction to achieve fairness is by\nlearning data representations that remove the semantics of protected\ncharacteristics, and are therefore able to mitigate unfair outcomes. All\navailable models however learn latent embeddings which comes at the cost of\nbeing uninterpretable. We propose to cast this problem as data-to-data\ntranslation, i.e. learning a mapping from an input domain to a fair target\ndomain, where a fairness definition is being enforced. Here the data domain can\nbe images, or any tabular data representation. This task would be\nstraightforward if we had fair target data available, but this is not the case.\nTo overcome this, we learn a highly unconstrained mapping by exploiting\nstatistics of residuals - the difference between input data and its translated\nversion - and the protected characteristics. When applied to the CelebA dataset\nof face images with gender attribute as the protected characteristic, our model\nenforces equality of opportunity by adjusting the eyes and lips regions.\nIntriguingly, on the same dataset we arrive at similar conclusions when using\nsemantic attribute representations of images for translation. On face images of\nthe recent DiF dataset, with the same gender attribute, our method adjusts nose\nregions. In the Adult income dataset, also with protected gender attribute, our\nmodel achieves equality of opportunity by, among others, obfuscating the wife\nand husband relationship. Analyzing those systematic changes will allow us to\nscrutinize the interplay of fairness criterion, chosen protected\ncharacteristics, and prediction performance.","url_abs":"http://arxiv.org/abs/1810.06755v2","url_pdf":"http://arxiv.org/pdf/1810.06755v2.pdf","source":{"archive":"pwc-archive (Hugging Face), CC BY-SA 4.0","snapshot":"2025-07-28","licence_url":"https://creativecommons.org/licenses/by-sa/4.0/legalcode","row_kind":"abstracts"},"code_links":[{"paper_slug":"neural-styling-for-interpretable-fair","repo_url":"https://github.com/predictive-analytics-lab/Data-Domain-Fairness","is_official":1,"mentioned_in_paper":1,"mentioned_in_github":0,"framework":"none","reach":null}],"tasks":[{"task_slug":"attribute","task_name":"Attribute"},{"task_slug":"fairness","task_name":"Fairness"},{"task_slug":"translation","task_name":"Translation"}],"methods":[],"datasets_introduced":[],"methods_introduced":[],"results":[],"syntology":{"atlas_url":null,"mcp":null,"developers":"https://syntology.ai/developers"},"arxiv_metadata":null,"syntology_extracted_results":null}