Research map / Learning theory and algorithms / Prediction and decision losses
Optimal Black-Box Reductions Between Optimization Objectives
Read the paperCheck your idea against 108,967 papers
AI-extracted summary. Written by AI from the paper's title only (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 |
|---|---|---|---|
| apply AdaptReg reduction (geometrically decreasing regularization) to SVRG for smooth non-SC objectives | SVRG · finite-sum smooth non-SC (Lasso, logistic regression) | running time complexity: removes the log(1/ε) factor multiplying n+L/ε compared with classical regularization | Helped |
| apply AdaptReg to Katyusha for smooth non-SC finite-sum objectives | Katyusha · finite-sum smooth non-SC | running time complexity: achieves 1/sqrt(ε) rate instead of log(1/ε)/sqrt(ε) | Helped |
| apply AdaptReg to APCG (Lasso experiments) rather than classical regularization with fixed sigma (ablation) | APCG · Lasso on covtype, mnist, rcv1 (9 settings of l1 weight) | objective distance to minimum vs passes: AdaptReg more efficient than old regularization reduction in all settings | Helped |
| apply AdaptReg to SDCA (Lasso experiments) (ablation) | SDCA · Lasso on covtype, mnist, rcv1 | objective distance to minimum vs passes | Helped |
| apply AdaptSmooth reduction (geometrically decreasing smoothing) to SVRG for SC non-smooth SVM (ablation) | SVRG · SVM on covtype, mnist, rcv1 | objective distance to minimum vs passes: especially when desired training error is small | Helped |
| apply AdaptSmooth to Katyusha for SC non-smooth finite-sum objectives | Katyusha · finite-sum SC non-smooth (SVM) | running time complexity: removes log(1/ε) factor, achieving 1/sqrt(ε) rate | Helped |
| classical regularization/smoothing reductions with fixed parameter vs adaptive parameter sequence (ablation) | AdaptReg and AdaptSmooth tuning · Lasso and SVM experiments | parameter tuning effort | No clear effect |
Closest papers by meaning
- Learning SMaLL Predictors (NeurIPS 2018 · similarity 0.87)
- Multiclass Loss Geometry Matters for Generalization of Gradient Descent in Separable Classification (NeurIPS 2025 · similarity 0.87)
- A Survey of Learning Criteria Going beyond the Usual Risk (Abstract Reprint) (AAAI 2024 · similarity 0.86)
- From Logistic Regression to the Perceptron Algorithm: Exploring Gradient Descent with Large Step Sizes (AAAI 2025 · similarity 0.85)
- On the Convergence of Loss and Uncertainty-based Active Learning Algorithms (NeurIPS 2024 · similarity 0.85)
- A Universal Growth Rate for Learning with Smooth Surrogate Losses (NeurIPS 2024 · similarity 0.85)
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.