Research map / Optimization / Nonconvex and smooth optimization
A Unified Formulation and Fast Accelerated Proximal Gradient Method for Classification
Naoki Ito, Akiko Takeda, Kim-Chuan Toh
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 bisection instead of breakpoint search for the projection (ablation) | FAPG · Projection, n=1,000,000, range [0,10]n | average runtime (msec): 94.2 -> 49.9 | Improved |
| use bisection instead of breakpoint search for the projection (ablation) | FAPG · Projection, n=100,000, range [0,10]n | average runtime (msec): 8.3 -> 4.9 | Improved |
| add the decreasing strategy (dec) to backtracking and restart (ablation) | FAPG nu-SVM · Artificial (m,n,nu)=(10000,1000,0.5) | convergence stability | Mixed |
| add the stabilizing strategy (st) on top of bt+re+dec (ablation) | FAPG nu-SVM · Artificial (m,n,nu)=(10000,1000,0.5) | convergence stability | Improved |
| add the final strategy (mt) on top of bt+re+dec+st (ablation) | FAPG nu-SVM · Artificial (m,n,nu)=(10000,1000,0.5) | optimality violation | Improved |
| add all acceleration strategies instead of exact Lipschitz constant (ablation) | FAPG nu-SVM · Artificial (m,n,nu)=(10000,1000,0.5) | constant L_k: 3.61x10^4 -> 3.68x10^2 | Improved |
| add 'mt' strategy (ablation) | FAPG logistic regression · a8a (C=10) | violation below 10^-6 within 10000 iterations | Improved |
| decrease xi from 10^-2 to 10^-3 in approximate logistic regression (ablation) | FAPG logistic regression · Artificial | accuracy: 99.8% -> 100.0% | Improved |
| decrease xi from 10^-2 to 10^-3 in approximate logistic regression (ablation) | FAPG logistic regression · mushrooms | accuracy: 99.4% -> 100.0% | Improved |
| use FAPG logistic regression with bias instead of LIBLINEAR without bias | FAPG · Artificial | accuracy: 85.2% -> 100.0% | Improved |
| use FAPG instead of LIBSVM and SeDuMi on large datasets (linear kernel) | FAPG nu-SVM · Benchmark datasets with m >= 50000 | computation time | Improved |
| use RBF kernel instead of linear kernel (ablation) | FAPG nu-SVM · Benchmark datasets | runtime | Hurt |
Closest papers by meaning
- Solving L1-regularized SVMs and Related Linear Programs: Revisiting the Effectiveness of Column and Constraint Generation (JMLR 2022 · similarity 0.84)
- Approximation Vector Machines for Large-scale Online Learning (JMLR 2017 · similarity 0.83)
- Fast Support Vector Classification for Large-Scale Problems (TPAMI 2021 · similarity 0.83)
- Adaptive Proximal Gradient Methods for Structured Neural Networks (NeurIPS 2021 · similarity 0.83)
- Faster Algorithms for Structured Linear and Kernel Support Vector Machines (ICLR 2025 · similarity 0.82)
- Principled Algorithms for Optimizing Generalized Metrics in Binary Classification (ICML 2025 · 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.
