Research map / Adversarial robustness and security / Certified and empirical robustness
Achieving Robustness in the Wild via Adversarial Mixing With Disentangled Representations
Check 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 adversarially selected StyleGAN latent mixes instead of nominal training | CelebA smile classifier · CelebA, Attribute #2 (smiling) test | accuracy: 90.22% -> 92.29% | Helped |
| Train with adversarially selected semantic variations instead of random latent mixes | AdvMix / RandMix · Color-MNIST, clean test, StyleGAN more biased | accuracy: 97.63% -> 98.79% | Helped |
| Train with adversarially selected semantic variations instead of random latent mixes | AdvMix / RandMix · RGB Color-MNIST, clean test, 99.9% red | accuracy: 53.56% -> 85.00% | Helped |
| Train with adversarially selected semantic variations instead of random latent mixes | AdvMix / RandMix · RGB Color-MNIST, clean test, unbiased | accuracy: 88.55% -> 85.07% | Hurt |
| Train with adversarially selected semantic variations instead of nominal training | CelebA classifiers · CelebA, attributes #1-#4 clean test | accuracy: 96.49% -> 97.56%; 90.22% -> 92.29%; 83.52% -> 85.65%; 78.05% -> 79.47% | Helped |
| Train with adversarially selected semantic variations instead of nominal training | CelebA classifiers · CelebA, attributes #1-#4 under perturbation | accuracy: 40.35% -> 84.29%; 18.60% -> 74.55%; 3.31% -> 69.55%; 0.23% -> 47.95% | Helped |
| Use ℓ∞ adversarial training instead of nominal training | CelebA classifiers · CelebA, clean test | accuracy: 90.22% -> 91.11%; 83.52% -> 81.43%; 78.05% -> 76.61% | Mixed |
| Use random latent mixing instead of nominal training | RandMix · CelebA clean test | accuracy: 96.49% -> 96.70%; 90.22% -> 90.36%; 83.52% -> 84.49%; 78.05% -> 76.41% | Mixed |
| Use CutMix instead of nominal training | CelebA classifiers · CelebA Attribute #3 | clean accuracy: 83.52% -> 85.67% | Helped |
| Use CutMix instead of nominal training | CelebA smile classifier · CelebA Attribute #2 (smiling), under perturbation | accuracy: 18.60% -> 15.46% | Hurt |
Closest papers by meaning
- Encoding Robustness to Image Style via Adversarial Feature Perturbations (NeurIPS 2021 · similarity 0.89)
- Attribute-Guided Adversarial Training for Robustness to Natural Perturbations (AAAI 2021 · similarity 0.88)
- What It Thinks Is Important Is Important: Robustness Transfers Through Input Gradients (CVPR 2020 · similarity 0.88)
- Mitigating Feature Gap for Adversarial Robustness by Feature Disentanglement (AAAI 2025 · similarity 0.88)
- Rethinking Invariance Regularization in Adversarial Training to Improve Robustness-Accuracy Trade-off (ICLR 2025 · similarity 0.87)
- CAT: Contrastive Adversarial Training for Evaluating the Robustness of Protective Perturbations in Latent Diffusion Models (ICML 2025 · 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.