Research map / Adversarial robustness and security / Attacks on text and graphs
Improving Adversarial Robustness with Self-Paced Hard-Class Pair Reweighting
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 self-paced hard/easy-class pair reweighting with its consistency regularizer | adversarial training on CIFAR-10 · CIFAR-10, white-box PGD-20 | accuracy: 54.87 -> 58.33 | Helped |
| use self-paced hard/easy-class pair reweighting with its consistency regularizer | adversarial training on MNIST · MNIST, white-box PGD-20 | accuracy: 96.24 -> 96.64 | Helped |
| add self-paced factors to the accuracy loss (ablation) | SPAT on CIFAR-10 · CIFAR-10, PGD-20 | accuracy: 58.74 -> 59.56 | Helped |
| add the self-paced HCP/ECP consistency term to the robustness loss (ablation) | SPAT on CIFAR-10 · CIFAR-10, PGD-20 | accuracy: 56.14 -> 59.56 | Helped |
| add self-paced factors to both accuracy and robustness losses (ablation) | NCE adversarial training on CIFAR-10 · CIFAR-10, PGD-20 | accuracy: 54.58 -> 59.56 | Helped |
| use NCE instead of self-paced CE (ablation) | SPAT on CIFAR-10 · CIFAR-10, PGD-20 | accuracy: 53.55 -> 59.56 | Helped |
| replace standard CE with self-paced CE (ablation) | adversarial training on CIFAR-10 · CIFAR-10, PGD-20 | accuracy: 51.82 -> 53.55 | Helped |
| increase embedding scale s from 1 to 10 (ablation) | SPAT on CIFAR-10 · CIFAR-10, PGD-20 | accuracy: 49.52 -> 61.02 | Helped |
| increase embedding scale s from 1 to 10 (ablation) | SPAT on CIFAR-10 · CIFAR-10 clean | accuracy: 87.57 -> 81.24 | Hurt |
| increase alpha from 0.0 to 0.6 (ablation) | SPAT on CIFAR-10 · CIFAR-10, PGD-20 | accuracy: 58.32 -> 60.57 | Helped |
| increase beta from 0.0 to 0.4 (ablation) | SPAT on CIFAR-10 · CIFAR-10, PGD-20 | accuracy: 57.88 -> 59.64 | Helped |
| increase beta from 0.4 to 0.6 (ablation) | SPAT on CIFAR-10 · CIFAR-10, PGD-20 | accuracy: 59.64 -> 57.62 | Hurt |
Closest papers by meaning
- Doubly Robust Instance-Reweighted Adversarial Training (ICLR 2024 · similarity 0.89)
- WAT: Improve the Worst-Class Robustness in Adversarial Training (AAAI 2023 · similarity 0.89)
- Adversarial Robustness Against the Union of Multiple Perturbation Models (ICML 2020 · similarity 0.88)
- Bag of Tricks for Adversarial Training (ICLR 2021 · similarity 0.88)
- Adversarial Training Can Provably Improve Robustness: Theoretical Analysis of Feature Learning Process Under Structured Data (ICLR 2025 · similarity 0.87)
- Clarifying the Behavior and the Difficulty of Adversarial Training (AAAI 2024 · similarity 0.87)
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.