Research map / Adversarial robustness and security / Certified and empirical robustness
Fast and Scalable Adversarial Training of Kernel SVM via Doubly Stochastic Gradients
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 |
|---|---|---|---|
| Use adversarially trained kernel SVM instead of natural DSG | Kernel SVM · CIFAR10 automobile vs. truck, FGSM | accuracy (%): 65.10$\pm$0.51 -> 74.65$\pm$0.53 | Helped |
| Use adversarially trained kernel SVM instead of natural DSG | Kernel SVM · CIFAR10 automobile vs. truck, PGD | accuracy (%): 63.32$\pm$0.47 -> 66.37$\pm$0.59 | Helped |
| Use adversarially trained kernel SVM instead of natural DSG | Kernel SVM · CIFAR10 automobile vs. truck, C&W | accuracy (%): 47.30$\pm$0.76 -> 51.76$\pm$0.81 | Helped |
| Use adversarially trained kernel SVM instead of natural DSG | Kernel SVM · MNIST8m 0 vs. 4, C&W | accuracy (%): 89.42$\pm$0.37 -> 95.46$\pm$0.49 | Helped |
| Use adversarially trained kernel SVM instead of natural DSG | Kernel SVM · MNIST8m 0 vs. 4, Normal | accuracy (%): 99.48$\pm$0.08 -> 99.46$\pm$0.10 | No clear effect |
| Use constant rather than diminishing stepsize (ablation) | adv-SVM · CIFAR10 automobile vs. truck, PGD | accuracy (%): 64.93$\pm$0.56 -> 66.37$\pm$0.59 | Helped |
| Use constant rather than diminishing stepsize (ablation) | adv-SVM · CIFAR10 automobile vs. truck, Normal | accuracy (%): 74.88$\pm$0.46 -> 75.22$\pm$0.41 | No clear effect |
| Use constant rather than diminishing stepsize (ablation) | adv-SVM · MNIST8m 0 vs. 4, C&W | accuracy (%): 93.87$\pm$0.44 -> 95.46$\pm$0.49 | Helped |
| Use constant rather than diminishing stepsize (ablation) | adv-SVM · MNIST8m 0 vs. 4, Normal | accuracy (%): 99.47$\pm$0.08 -> 99.46$\pm$0.10 | No clear effect |
| Train the kernel SVM with adversarial training rather than linear-SVM adversarial training | SVM · CIFAR10 automobile vs. truck, C&W | accuracy (%): 48.82$\pm$0.93 -> 51.76$\pm$0.81 | Helped |
| Train the kernel SVM with adversarial training rather than linear-SVM adversarial training | SVM · MNIST8m 0 vs. 4, ZOO | accuracy (%): 88.39$\pm$0.75 -> 95.32$\pm$0.56 | Helped |
Closest papers by meaning
- Kernel Learning with Adversarial Features: Numerical Efficiency and Adaptive Regularization (NeurIPS 2025 · similarity 0.90)
- Adversarial Training Can Provably Improve Robustness: Theoretical Analysis of Feature Learning Process Under Structured Data (ICLR 2025 · similarity 0.89)
- Fast is better than free: Revisiting adversarial training (ICLR 2020 · similarity 0.89)
- Taxonomy Driven Fast Adversarial Training (AAAI 2024 · similarity 0.87)
- Robust Models are less Over-Confident (NeurIPS 2022 · similarity 0.87)
- Ensemble Adversarial Training: Attacks and Defenses (ICLR 2018 · similarity 0.86)
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.