Research map / Adversarial robustness and security / Certified and empirical robustness
Cost-Sensitive Robustness against Adversarial Examples
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 |
|---|---|---|---|
| Train with cost weights for the selected transformation instead of treating all class transformations equally | MNIST classifier · MNIST, single-pair tasks (0,2), (6,5), and (4,9) | cost-sensitive robust error: 0.92% -> 0.31%; 3.55% -> 0.42%; 10.08% -> 1.02% | Helped |
| Train with cost weights for transformations from a selected seed class | MNIST classifier · MNIST, single-seed tasks | cost-sensitive robust error: 3.67% -> 0.92%; 14.34% -> 3.68%; 22.28% -> 5.75% | Helped |
| Train with cost weights for transformations to a selected target class | MNIST classifier · MNIST, single-target tasks | cost-sensitive robust error: 2.23% -> 0.14%; 3.10% -> 0.29%; 5.24% -> 0.54% | Helped |
| Use cost-sensitive weights for selected multiple transformations | MNIST classifier · MNIST, multiple-transformation tasks | cost-sensitive robust error: 11.14% -> 7.02%; 5.01% -> 2.18%; 14.45% -> 9.97%; 13.13% -> 9.44% | Helped |
| Train with cost weights for valued transformations | MNIST classifier · MNIST, small-large real-valued costs | robust cost: 2.245 -> 0.947 | Helped |
| Train with cost weights for valued transformations | MNIST classifier · MNIST, large-small real-valued costs | robust cost: 3.344 -> 1.549 | Helped |
| Train with vehicle-seed cost weights, making vehicle-to-animal transformations more costly | CIFAR10 classifier · CIFAR10, vehicle real-valued task | robust cost: 4.183 -> 3.095 | Helped |
| Use cost-sensitive robust training for the selected pair | CIFAR10 classifier · CIFAR10, frog→bird and cat→plane | cost-sensitive robust error: 19.90% -> 1.20%; 9.30% -> 2.60% | Helped |
| Use cost-sensitive robust training for selected seeds, targets, and cross-group transformations | CIFAR10 classifier · CIFAR10 single-seed, single-target, and animal/vehicle tasks | cost-sensitive robust error: 57.20% -> 28.90%; 35.60% -> 15.40%; 16.99% -> 3.77%; 9.42% -> 3.06%; 16.67% -> 7.42%; 12.07% -> 8.00% | Helped |
| Increase the robustness regularization parameter α | Cost-sensitive robust model · MNIST single-seed task, digit 9 | cost-sensitive robust error | Helped |
| Increase the robustness regularization parameter α | Cost-sensitive robust model · MNIST single-seed task, digit 9 | classification error | Hurt |
| Use standard non-adversarial cost-sensitive training rather than cost-sensitive robust training | MNIST classifier · MNIST, small-large real-valued task | robust cost: 85.197 -> 85.197 | No clear effect |
Closest papers by meaning
- Provably Cost-Sensitive Adversarial Defense via Randomized Smoothing (ICML 2025 · similarity 0.94)
- Certified Adversarial Robustness with Additive Noise (NeurIPS 2019 · similarity 0.90)
- Theoretically Principled Trade-off between Robustness and Accuracy (ICML 2019 · similarity 0.90)
- On the Sensitivity of Adversarial Robustness to Input Data Distributions (ICLR 2019 · similarity 0.90)
- Robustness May Be at Odds with Accuracy (ICLR 2019 · similarity 0.90)
- Adversarial Training Can Provably Improve Robustness: Theoretical Analysis of Feature Learning Process Under Structured Data (ICLR 2025 · similarity 0.88)
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.