Research map / Label-efficient and robust learning / Recognition and long-tail
DDSVM: A Differentiable Framework for Deep Support Vector Machines with Iterative Geometry-Aware Optimization
Yirun Ding, Zhihui Lai
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 |
|---|---|---|---|
| Train with DDSVM instead of CE in the 8% data regime | ResNet-18 · CIFAR-10 (Reduced) | Top-1 error (%): 30.75 -> 23.21 | Improved |
| Train with DDSVM instead of CE with full data | ResNet-18 · CIFAR-10/100, SVHN (Full) | Top-1 error (%): 4.64 -> 4.70 (CIFAR-10) | No clear effect |
| DDSVM on SVHN reduced data | ResNet-18 · SVHN (Reduced) | Top-1 error (%): 6.08 -> 6.79 | Hurt |
| DDSVM on STL-10 full data | ResNet-18 · STL-10 (Full) | Top-1 error (%): 19.45 -> 17.43 | Improved |
| Add Pure Squeeze term to CE (ablation) | DDSVM · CIFAR-10 (Reduced) | Top-1 error (%): 30.75 -> 24.81 | Improved |
| Add Rest Class Push/Repulsion (ablation) | DDSVM · CIFAR-10 (Reduced) | Top-1 error (%): 24.81 -> 23.88 | Improved |
| Add GAS perturbation (eps_adv 0 to 1.0) (ablation) | DDSVM · CIFAR-10 (Reduced) | Top-1 error (%): 23.88 -> 23.21 | Improved |
| Increase GAS strength to 10.0 vs 1.0 (ablation) | DDSVM · CIFAR-10 (Reduced) | Top-1 error (%): 23.21 -> 23.27 | No clear effect |
| Combine DDSVM with Mixup | DDSVM · CIFAR-10 (Reduced) | Top-1 error (%): 23.21 -> 17.67 | Improved |
| Sweep beta, p, theta, SVM C over 50x ranges (ablation) | DDSVM · CIFAR-10 (Reduced) | Top-1 error (%): 23.75-24.01 | No clear effect |
| Replace OVR SVM with Cramer-Singer SVM (ablation) | DDSVM · CIFAR-100 (Reduced) | Top-1 error (%): 23.06 -> 23.11 | No clear effect |
| DDSVM vs CE on corrupted CIFAR-10 | ResNet-18 · CIFAR-10-C severity 5 | mCE (%): 42.25 -> 41.79 | Mixed |
Earlier methods this paper compares with
Named baselines from earlier accepted papers in this paper's results tables: better than 2 of 2 on most shared results.
- ArcFace — ArcFace: Additive Angular Margin Loss for Deep Face Recognition (CVPR 2019) · better on 1 of 1 shared result
- CosFace — CosFace: Large Margin Cosine Loss for Deep Face Recognition (CVPR 2018) · better on 1 of 1 shared result
Matched by method name in results tables read by AI. Authors' own tables tend to favour the newer method, and a comparison can be missed when a method is renamed.
Closest papers by meaning
- Deep Support Vectors (NeurIPS 2024 · similarity 0.84)
- Solving L1-regularized SVMs and Related Linear Programs: Revisiting the Effectiveness of Column and Constraint Generation (JMLR 2022 · similarity 0.84)
- Deep Principal Support Vector Machines for Nonlinear Sufficient Dimension Reduction (ICML 2025 · similarity 0.83)
- Local Duality for Sparse Support Vector Machines (TPAMI 2026 · similarity 0.83)
- Faster Algorithms for Structured Linear and Kernel Support Vector Machines (ICLR 2025 · similarity 0.82)
- Differentiable Model Scaling using Differentiable Topk (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.
