Research map / Learning theory and algorithms / Fairness
Exploiting MMD and Sinkhorn Divergences for Fair and Transferable Representation Learning
Luca Oneto, Michele Donini, Giulia Luise, Carlo Ciliberto, Andreas Maurer, Massimiliano Pontil
Read the paperCheck your idea against 137,217 papers
AI-extracted. The measured effects below were read by AI from the paper's full text (no summary yet). Numbers and claims can be wrong or missing context, so check the paper before you cite or build on it.
Measured effects reported in the paper
| What was changed | Where | Measured | Effect |
|---|---|---|---|
| add Sinkhorn-divergence fairness constraint on shared representation | shared-representation FFNN · School new tasks (linear) | DDP: .088 -> .017 | Improved |
| add Sinkhorn constraint, unconstrained model as reference | shared-representation FFNN · School new tasks (linear) | ERR: 13.8 -> 14.8 | Hurt |
| replace M [13] with Sinkhorn-constrained representation | shared-representation FFNN · UNIV new tasks (linear) | DDP: .033 -> .009 | Improved |
| replace M [13] with Sinkhorn-constrained representation | shared-representation FFNN · UNIV new tasks (linear) | ERR: 27.3 -> 16.0 | Improved |
| use Gaussian-kernel MMD instead of mean matching (ablation) | shared-representation FFNN · School new tasks (linear) | DDP: .022 -> .020 | Improved |
| use Sinkhorn instead of mean matching (ablation) | shared-representation FFNN · UNIV new tasks (linear) | DDP: .015 -> .009 | Improved |
| use Sinkhorn instead of mean matching (ablation) | shared-representation FFNN · Movielens new tasks (linear) | DDP: .016 -> .011 | Improved |
| replace M [12] output-level constraint with Sinkhorn representation constraint | shared-representation FFNN · Movielens new tasks (linear) | DDP: .025 -> .011 | Improved |
| replace linear with sigmoid activation in representation (ablation) | shared-representation FFNN · School new tasks | ERR: 14.8 -> 10.2 | Improved |
| add Sinkhorn fairness constraint with sigmoid representation | shared-representation FFNN · UNIV new tasks | DDP: .175 -> .115 | Mixed |
| replace Sinkhorn with M [13] on non-linear model | shared-representation FFNN · UNIV new tasks (sigmoid) | DDP: .101 -> .115 | Hurt |
| fairness constraint [12] on output with sensitive feature in model | M [12] · UNIV new tasks (sigmoid) | DDP: .212 -> .241 | Hurt |
Closest papers by meaning
- Utility-Fairness Trade-Offs and How to Find Them (CVPR 2024 · similarity 0.88)
- Fair Representation Learning with Controllable High Confidence Guarantees via Adversarial Inference (NeurIPS 2025 · similarity 0.88)
- On Optimal Steering to Achieve Exact Fairness (NeurIPS 2025 · similarity 0.88)
- Rethinking Fair Representation Learning for Performance-Sensitive Tasks (ICLR 2025 · similarity 0.87)
- Geometry of Relaxed Fair Regression: A Unified Framework for Aware and Unaware Settings (NeurIPS 2026 · similarity 0.87)
- FairMT: Fairness for Heterogeneous Multi-Task Learning (NeurIPS 2026 · similarity 0.87)
Writing a paper in this area? scime draft is a free LaTeX editor with AI included, real-time co-authors and a check of your references and venue rules before you submit. It also imports OpenAI Prism projects.
