Research map / Optimization / Optimizers
Implicit rate-constrained optimization of non-decomposable objectives
Abhishek Kumar, Harikrishna Narasimhan, Andrew Cotter
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 |
|---|---|---|---|
| Optimize FNR directly while enforcing the target FPR using implicit thresholds rather than cross-entropy training | CelebA attribute classification · High-cheekbones, FPR 1% | FNR (%): 53.5 -> 46.9 | Improved |
| Use implicit thresholds instead of TFCO's constrained optimizer | CelebA attribute classification · High-cheekbones, FPR 5% | FNR (%): 30.1 -> 28.5 | Improved |
| Optimize partial ROC-AUC with ICO instead of cross-entropy | CelebA image classification · Mean over 8 attributes, FPR range [0,5%] | partial ROC-AUC: 74.7 -> 76.8 | Improved |
| Optimize partial ROC-AUC with ICO instead of TFCO | BigEarthNet image classification · Mean over 5 labels, FPR range [0,5%] | partial ROC-AUC: 66.2 -> 68.5 | Improved |
| Optimize high-recall PR-AUC directly with ICO instead of cross-entropy | UCI Letter · Recall [0.95, 1] | PR-AUC: 15.13 ± 0.86 -> 23.04 ± 0.77 | Improved |
| Optimize high-recall PR-AUC directly with ICO instead of TFCO | UCI IJCNN1 · Recall [0.95, 1] | PR-AUC: 26.14 ± 0.57 -> 27.28 ± 0.58 | Improved |
| Use ICO instead of TFCO for high-recall PR-AUC | UCI Communities & Crime · Recall [0.95, 1] | PR-AUC: 47.04 ± 0.97 -> 47.03 ± 1.08 | No clear effect |
| Use ICO rather than TFCO for partial ROC-AUC | CelebA image classification · Mean over 8 attributes, FPR range [0,1%] | partial ROC-AUC: 65.9 -> 66.1 | Mixed |
Closest papers by meaning
- Optimizing Generalized Rate Metrics with Three Players (NeurIPS 2019 · similarity 0.87)
- Optimization with Non-Differentiable Constraints with Applications to Fairness, Recall, Churn, and Other Goals (JMLR 2019 · similarity 0.85)
- Near-Optimal Solutions of Constrained Learning Problems (ICLR 2024 · similarity 0.83)
- Constrained Bi-Level Optimization: Proximal Lagrangian Value Function Approach and Hessian-free Algorithm (ICLR 2024 · similarity 0.83)
- Smoothing Proximal Gradient Methods for Nonsmooth Sparsity Constrained Optimization: Optimality Conditions and Global Convergence (ICML 2024 · similarity 0.83)
- A Universal Class of Sharpness-Aware Minimization Algorithms (ICML 2024 · similarity 0.82)
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.
