Papers by Somnath Basu Roy Chowdhury
Learning Fair Representations via Rate-Distortion Maximization (2022.tacl-1)
Copied to clipboard
| Challenge: | Empirical evaluations show that FaRM debiases representations with or without a target task at hand. |
| Approach: | They propose a method that makes representations belonging to the same protected attribute class uncorrelated, using the rate-distortion function. |
| Outcome: | Empirical results show that the proposed technique achieves state-of-the-art performance on several datasets and leaks significantly less protected attribute information against an attack by a non-linear probing network. |