{"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/auditing-black-box-models-for-indirect","title":"Auditing Black-box Models for Indirect Influence","arxiv_id":"1602.07043","date":"2016-02-23","proceeding":null,"authors":["Philip Adler","Casey Falk","Sorelle A. Friedler","Gabriel Rybeck","Carlos Scheidegger","Brandon Smith","Suresh Venkatasubramanian"],"abstract":"Data-trained predictive models see widespread use, but for the most part they\nare used as black boxes which output a prediction or score. It is therefore\nhard to acquire a deeper understanding of model behavior, and in particular how\ndifferent features influence the model prediction. This is important when\ninterpreting the behavior of complex models, or asserting that certain\nproblematic attributes (like race or gender) are not unduly influencing\ndecisions.\n  In this paper, we present a technique for auditing black-box models, which\nlets us study the extent to which existing models take advantage of particular\nfeatures in the dataset, without knowing how the models work. Our work focuses\non the problem of indirect influence: how some features might indirectly\ninfluence outcomes via other, related features. As a result, we can find\nattribute influences even in cases where, upon further direct examination of\nthe model, the attribute is not referred to by the model at all.\n  Our approach does not require the black-box model to be retrained. This is\nimportant if (for example) the model is only accessible via an API, and\ncontrasts our work with other methods that investigate feature influence like\nfeature selection. We present experimental evidence for the effectiveness of\nour procedure using a variety of publicly available datasets and models. We\nalso validate our procedure using techniques from interpretable learning and\nfeature selection, as well as against other black-box auditing procedures.","url_abs":"http://arxiv.org/abs/1602.07043v2","url_pdf":"http://arxiv.org/pdf/1602.07043v2.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":"auditing-black-box-models-for-indirect","repo_url":"https://github.com/cfalk/BlackBoxAuditing","is_official":1,"mentioned_in_paper":1,"mentioned_in_github":1,"framework":"tf","reach":{"status":"unanswered"}},{"paper_slug":"auditing-black-box-models-for-indirect","repo_url":"https://github.com/algofairness/BlackBoxAuditing","is_official":0,"mentioned_in_paper":0,"mentioned_in_github":1,"framework":"tf","reach":{"status":"unanswered"}}],"tasks":[{"task_slug":"attribute","task_name":"Attribute"},{"task_slug":"feature-selection","task_name":"feature selection"}],"methods":[],"datasets_introduced":[],"methods_introduced":[],"results":[],"syntology":{"atlas_url":"https://app.syntology.ai/?focus=1602.07043","mcp":null,"developers":"https://syntology.ai/developers"},"arxiv_metadata":null,"syntology_extracted_results":null}