Research map / Adversarial robustness and security / Certified and empirical robustness
On the Global Optima of Kernelized Adversarial Representation Learning
Bashir Sadeghi, Runyi Yu, Vishnu Boddeti
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 closed-form Spectral-ARL (Linear-SARL) instead of DNN ML-ARL | ARL encoder · Adult | target accuracy, income (%): 84.4 -> 84.1 | Hurt |
| use Linear-SARL instead of MaxEnt-ARL | ARL encoder · German | target accuracy, credit (%): 72.5 -> 76.3 | Improved |
| use Linear-SARL instead of ML-ARL | ARL encoder · German | target accuracy, credit (%): 74.4 -> 76.3 | Improved |
| use Kernel-SARL instead of Linear-SARL (ablation) | SARL · Adult and German | target and sensitive accuracy (%): 84.1 -> 84.1 | No clear effect |
| use Linear-SARL instead of MaxEnt-ARL | ARL encoder · Extended Yale B | adversary accuracy, illumination (%): 40 -> 21 | Improved |
| use Kernel-SARL instead of MaxEnt-ARL | ARL encoder · Extended Yale B | target accuracy, identity (%): 89 -> 86 | Hurt |
| use Kernel-SARL instead of Linear-SARL (ablation) | SARL · Extended Yale B | target accuracy, identity (%): 81 -> 86 | Improved |
| exchange training and test sets (ablation) | Linear-SARL · Extended Yale B | target accuracy, identity (%): 81 -> 86 | Improved |
| use kernel (non-linear) encoder instead of linear encoder (ablation) | SARL · Gaussian mixture | adversary accuracy (%): 50% | Improved |
| use kernel encoder instead of linear encoder (ablation) | SARL · CIFAR-100 | target vs adversary trade-off | Improved |
| use closed-form SARL instead of simultaneous SGD | ARL encoder · CIFAR-100 | training stability | Improved |
Closest papers by meaning
- Kernel Learning with Adversarial Features: Numerical Efficiency and Adaptive Regularization (NeurIPS 2025 · similarity 0.90)
- Controllable Invariance through Adversarial Feature Learning (NeurIPS 2017 · similarity 0.90)
- Invariant Representations without Adversarial Training (NeurIPS 2018 · similarity 0.90)
- Convex Representation Learning for Generalized Invariance in Semi-Inner-Product Space (ICML 2020 · similarity 0.88)
- Rethinking Invariance Regularization in Adversarial Training to Improve Robustness-Accuracy Trade-off (ICLR 2025 · similarity 0.88)
- Unsupervised Adversarial Invariance (NeurIPS 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.
