Research map / Adversarial robustness and security / Certified and empirical robustness
Improved Network Robustness with Adversary Critic
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 the adversary critic objective instead of adversarial training | MNIST fully connected network · MNIST | robustness ρ (proposed attack): 0.299 -> 0.575 | Helped |
| Train with the adversary critic objective instead of adversarial training | MNIST LeNet-5 · MNIST | robustness ρ (proposed attack): 0.286 -> 0.590 | Helped |
| Train with the adversary critic objective instead of margin maximization | MNIST fully connected network · MNIST | robustness ρ (proposed attack): 0.453 -> 0.575 | Helped |
| Train with the adversary critic objective instead of margin maximization | MNIST LeNet-5 · MNIST | robustness ρ (proposed attack): 0.470 -> 0.590 | Helped |
| Use the adversary critic objective | MNIST fully connected network · MNIST | test error (%): 0.84 -> 1.18 | Hurt |
| Use the adversary critic objective | MNIST LeNet-5 · MNIST | test error (%): 0.54 -> 0.93 | Hurt |
| Use the adversary critic objective rather than adversarial training | MNIST fully connected network · MNIST human annotation experiment | adversarial images labeled as target (%): 19.02 -> 87.99 | Helped |
| Use the adversary critic objective rather than adversarial training | MNIST LeNet-5 · MNIST human annotation experiment | adversarial images labeled as target (%): 19.1 -> 92.29 | Helped |
| Select the target class with the closest decision boundary instead of sampling a masked uniform target | Adversary Critic · MNIST | robustness | Helped |
| Add adversarial cycle-consistency constraint | Adversary Critic · MNIST | robustness | Helped |
| Use class labels only in the critic's final classification layer rather than at each layer | Adversary Critic · MNIST | robustness | Helped |
Closest papers by meaning
- Improving Adversarial Robustness Requires Revisiting Misclassified Examples (ICLR 2020 · similarity 0.91)
- A Direct Approach to Robust Deep Learning Using Adversarial Networks (ICLR 2019 · similarity 0.91)
- Adversarial Training Can Provably Improve Robustness: Theoretical Analysis of Feature Learning Process Under Structured Data (ICLR 2025 · similarity 0.91)
- Towards Deep Learning Models Resistant to Adversarial Attacks (ICLR 2018 · similarity 0.90)
- Benchmarking Adversarial Robustness on Image Classification (CVPR 2020 · similarity 0.90)
- Wide Two-Layer Networks can Learn from Adversarial Perturbations (NeurIPS 2024 · 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.