{"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/fairness-without-demographics-in-repeated","title":"Fairness Without Demographics in Repeated Loss Minimization","arxiv_id":"1806.08010","date":"2018-06-20","proceeding":"ICML 2018 7","authors":["Tatsunori B. Hashimoto","Megha Srivastava","Hongseok Namkoong","Percy Liang"],"abstract":"Machine learning models (e.g., speech recognizers) are usually trained to\nminimize average loss, which results in representation disparity---minority\ngroups (e.g., non-native speakers) contribute less to the training objective\nand thus tend to suffer higher loss. Worse, as model accuracy affects user\nretention, a minority group can shrink over time. In this paper, we first show\nthat the status quo of empirical risk minimization (ERM) amplifies\nrepresentation disparity over time, which can even make initially fair models\nunfair. To mitigate this, we develop an approach based on distributionally\nrobust optimization (DRO), which minimizes the worst case risk over all\ndistributions close to the empirical distribution. We prove that this approach\ncontrols the risk of the minority group at each time step, in the spirit of\nRawlsian distributive justice, while remaining oblivious to the identity of the\ngroups. We demonstrate that DRO prevents disparity amplification on examples\nwhere ERM fails, and show improvements in minority group user satisfaction in a\nreal-world text autocomplete task.","url_abs":"http://arxiv.org/abs/1806.08010v2","url_pdf":"http://arxiv.org/pdf/1806.08010v2.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":"fairness-without-demographics-in-repeated","repo_url":"https://worksheets.codalab.org/worksheets/0x17a501d37bbe49279b0c70ae10813f4c","is_official":1,"mentioned_in_paper":0,"mentioned_in_github":0,"framework":"none","reach":null}],"tasks":[{"task_slug":"fairness","task_name":"Fairness"}],"methods":[],"datasets_introduced":[],"methods_introduced":[],"results":[],"syntology":{"syntology_url":null,"atlas_url":"https://app.syntology.ai/?focus=1806.08010","mcp":null,"developers":"https://syntology.ai/developers"},"arxiv_metadata":null,"syntology_extracted_results":null}