Research map / Learning theory and algorithms / Prediction and decision losses
Hinge-Minimax Learner for the Ensemble of Hyperplanes
Dolev Raviv, Tamir Hazan, Margarita Osadchy
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 |
|---|---|---|---|
| increase hyperplanes K from 1 to 4 (ablation) | KHHM · UCI Letter | classification rate at EER: 89.32 -> 94.48 | Improved |
| use KHHM K=4 instead of linear SVM | KHHM · UCI Letter | classification rate at EER: 84.87 -> 94.48 | Improved |
| use RBF kernel SVM instead of KHHM K=4 | KHHM · UCI Letter | classification rate at EER: 94.48 -> 96.47 | Improved |
| add a second hyperplane (K=2 vs K=1) (ablation) | KHHM · SUN 397 | AUC: 88.89 -> 90.99 | Improved |
| use more than two hyperplanes (ablation) | KHHM · SUN 397 | AUC | No clear effect |
| vary positive fraction from 0.01 to 0.5 (ablation) | KHHM K=5 · Synthetic 2D | AUC: 94.68 -> 95.83 | No clear effect |
| add latent components (LHM) instead of single-component KHHM (ablation) | LHM · PASCAL VOC 2007 | accuracy at EER: 69.45% -> 71.48% | Improved |
| jointly learn components instead of k-means plus LDA union | LHM · PASCAL VOC 2007 | accuracy at EER: 65.17% -> 71.48% | Improved |
| use hinge-minimax LHM instead of equivalent NN | LHM · PASCAL VOC 2007 | accuracy at EER: 67.19% -> 71.48% | Improved |
| transfer learning with LHM instead of CNN trained from scratch with few samples | LHM-NN · CIFAR-100 5 classes | accuracy | Improved |
| use histogram intersection kernel with 6300-dim features instead of KHHM K=2 | KHHM · SUN 397 | AUC: 92.99 vs 92.85 | No clear effect |
Closest papers by meaning
- On the Error Resistance of Hinge-Loss Minimization (NeurIPS 2020 · similarity 0.87)
- Best Fitting Hyperplanes for Classification (TPAMI 2016 · similarity 0.86)
- Solving L1-regularized SVMs and Related Linear Programs: Revisiting the Effectiveness of Column and Constraint Generation (JMLR 2022 · similarity 0.85)
- Theoretical Analysis of Adversarial Learning: A Minimax Approach (NeurIPS 2019 · similarity 0.85)
- The optimality of kernel classifiers in Sobolev space (ICLR 2024 · similarity 0.85)
- Multiclass Loss Geometry Matters for Generalization of Gradient Descent in Separable Classification (NeurIPS 2025 · similarity 0.84)
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.
