Reproduce “Doubly-Regressing Approach for Subgroup Fairness” on OpenResearch
Turn your favorite coding agents into research agents that review literature, develop hypotheses, and run experiments.
Universal app for Apple silicon and Intel. Requires macOS 11 or later.
About this paper
Kunwoong Kim, Kyungseon Lee, Jihu Lee, Dongyoon Yang, Yongdai Kim
Researchers from Seoul National University and SK Hynix Inc. developed the Doubly Regressing Adversarial learning for Fairness (DRAF) algorithm, a method that addresses data sparsity and computational challenges in intersectional subgroup fairness by introducing a novel surrogate measure, the Doubly Regressing R² (DR²). This approach achieves strong distributional fairness for both intersectional subgroups and marginal groups, demonstrating superior performance in sparse data environments and enabling more scalable fairness guarantees for complex AI systems.
Key insights
- The estimation error of empirical supIPM depends on the minimum sample size of subgroup-subsets, confirming that fairness can only be reliably guaranteed for sufficiently large, "active" subgroups.
- A theoretical link shows that DR² effectively approximates IPM for a single subgroup-subset, enabling the DR gap to serve as an upper bound for supIPM and significantly reduce computational complexity.
- Explicitly including marginal subgroups in the fairness objective is crucial to prevent degradation of marginal fairness, even when subgroup fairness is otherwise maintained.
Summary by alphaXiv.
Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.