Research map / Learning theory and algorithms / Submodular and game algorithms
Optimizing Generalized Rate Metrics with Three Players
Harikrishna Narasimhan, Andrew Cotter, Maya Gupta
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 |
|---|---|---|---|
| Use Alg. 2 stochastic classifier instead of oracle-based COCO | Alg. 2 · KL fairness under error constraint (Crime, Law, Adult, Wiki, COMPAS) | KL fairness metric: Crime 0.252 -> 0.120; Law 0.043 -> 0.054; Adult 0.011 -> 0.014; Wiki 0.134 -> 0.133 | Mixed |
| Use Alg. 2 instead of PostShift thresholding | Alg. 2 · KL fairness under error constraint | error-rate constraint satisfaction: Crime error ratio 1.40 -> 1.11 | Improved |
| Use Alg. 3 stochastic classifier instead of UncF1 | Alg. 3 · F-measure parity constraint | constraint violation: Crime 0.19 -> 0.11; Wiki 0.18 -> 0.00 | Improved |
| Use Alg. 3 stochastic classifier instead of UncF1 | Alg. 3 · F-measure parity constraint | overall F-measure: Wiki 0.967 -> 0.826; Law 0.975 -> 0.927 | Hurt |
| Use best deterministic iterate instead of stochastic classifier (ablation) | Alg. 3 · F-measure parity constraint (Law, Wiki) | constraint violation: Law 0.04 -> -0.05; Wiki 0.00 -> -0.05 | Improved |
| Use best deterministic iterate instead of stochastic classifier (ablation) | Alg. 3 · F-measure parity constraint (Law, Wiki) | overall F-measure: Law 0.927 -> 0.842; Wiki 0.826 -> 0.782 | Hurt |
| Use best deterministic iterate instead of stochastic classifier (ablation) | Alg. 2 · KL fairness under error constraint | KL fairness metric: Crime 0.120 -> 0.146; Wiki 0.133 -> 0.127 | No clear effect |
Closest papers by meaning
- Training Well-Generalizing Classifiers for Fairness Metrics and Other Data-Dependent Constraints (ICML 2019 · similarity 0.90)
- Accuracy vs. Accuracy: Computational Tradeoffs Between Classification Rates and Utility (NeurIPS 2026 · similarity 0.85)
- Principled Algorithms for Optimizing Generalized Metrics in Binary Classification (ICML 2025 · similarity 0.84)
- A General Online Algorithm for Optimizing Complex Performance Metrics (ICML 2024 · similarity 0.84)
- Near-Optimal Solutions of Constrained Learning Problems (ICLR 2024 · similarity 0.84)
- Monotone Individual Fairness (ICML 2024 · similarity 0.83)
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.
