Learning Fair Representations via Rate-Distortion Maximization
AbstractText representations learned by machine learning models often encode undesirable demographic information of the user. Predictive models based on these representations can rely on such information, resulting in biased decisions. We present a novel debiasing technique, Fairness...
| Main Authors: | , |
|---|---|
| Format: | Article |
| Language: | English |
| Published: |
The MIT Press
2022-01-01
|
| Series: | Transactions of the Association for Computational Linguistics |
| Online Access: | https://direct.mit.edu/tacl/article/doi/10.1162/tacl_a_00512/113492/Learning-Fair-Representations-via-Rate-Distortion |
| _version_ | 1828729130555801600 |
|---|---|
| author | Somnath Basu Roy Chowdhury Snigdha Chaturvedi |
| author_facet | Somnath Basu Roy Chowdhury Snigdha Chaturvedi |
| author_sort | Somnath Basu Roy Chowdhury |
| collection | DOAJ |
| description |
AbstractText representations learned by machine learning models often encode undesirable demographic information of the user. Predictive models based on these representations can rely on such information, resulting in biased decisions. We present a novel debiasing technique, Fairness-aware Rate Maximization (FaRM), that removes protected information by making representations of instances belonging to the same protected attribute class uncorrelated, using the rate-distortion function. FaRM is able to debias representations with or without a target task at hand. FaRM can also be adapted to remove information about multiple protected attributes simultaneously. Empirical evaluations show that FaRM achieves state-of-the-art performance on several datasets, and learned representations leak significantly less protected attribute information against an attack by a non-linear probing network. |
| first_indexed | 2024-04-12T14:28:43Z |
| format | Article |
| id | doaj.art-7657dd6258354bb58ee762fbd38e456e |
| institution | Directory Open Access Journal |
| issn | 2307-387X |
| language | English |
| last_indexed | 2024-04-12T14:28:43Z |
| publishDate | 2022-01-01 |
| publisher | The MIT Press |
| record_format | Article |
| series | Transactions of the Association for Computational Linguistics |
| spelling | doaj.art-7657dd6258354bb58ee762fbd38e456e2022-12-22T03:29:22ZengThe MIT PressTransactions of the Association for Computational Linguistics2307-387X2022-01-01101159117410.1162/tacl_a_00512Learning Fair Representations via Rate-Distortion MaximizationSomnath Basu Roy Chowdhury0Snigdha Chaturvedi1UNC Chapel Hill, USA somnath@cs.unc.eduUNC Chapel Hill, USA snigdha@cs.unc.edu AbstractText representations learned by machine learning models often encode undesirable demographic information of the user. Predictive models based on these representations can rely on such information, resulting in biased decisions. We present a novel debiasing technique, Fairness-aware Rate Maximization (FaRM), that removes protected information by making representations of instances belonging to the same protected attribute class uncorrelated, using the rate-distortion function. FaRM is able to debias representations with or without a target task at hand. FaRM can also be adapted to remove information about multiple protected attributes simultaneously. Empirical evaluations show that FaRM achieves state-of-the-art performance on several datasets, and learned representations leak significantly less protected attribute information against an attack by a non-linear probing network.https://direct.mit.edu/tacl/article/doi/10.1162/tacl_a_00512/113492/Learning-Fair-Representations-via-Rate-Distortion |
| spellingShingle | Somnath Basu Roy Chowdhury Snigdha Chaturvedi Learning Fair Representations via Rate-Distortion Maximization Transactions of the Association for Computational Linguistics |
| title | Learning Fair Representations via Rate-Distortion Maximization |
| title_full | Learning Fair Representations via Rate-Distortion Maximization |
| title_fullStr | Learning Fair Representations via Rate-Distortion Maximization |
| title_full_unstemmed | Learning Fair Representations via Rate-Distortion Maximization |
| title_short | Learning Fair Representations via Rate-Distortion Maximization |
| title_sort | learning fair representations via rate distortion maximization |
| url | https://direct.mit.edu/tacl/article/doi/10.1162/tacl_a_00512/113492/Learning-Fair-Representations-via-Rate-Distortion |
| work_keys_str_mv | AT somnathbasuroychowdhury learningfairrepresentationsviaratedistortionmaximization AT snigdhachaturvedi learningfairrepresentationsviaratedistortionmaximization |