Research map / Adversarial robustness and security / Certified and empirical robustness
LAFEAT: Piercing Through Adversarial Defenses With Latent Features
Yunrui Yu, Xitong Gao, Cheng-Zhong Xu
Check 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 LAFEAT MT(1k,10) instead of AutoAttack AA100 | Adversarial weight perturbation defense · CIFAR-10 | accuracy under attack (%): 60.04 -> 59.94 | Hurt |
| Use LAFEAT MT(1k,10) instead of AutoAttack AA100 | Manifold regularization defense · CIFAR-10 | accuracy under attack (%): 1.35 -> 0.21 | Hurt |
| Use LAFEAT MT(1k,10) instead of AutoAttack AA100 | Adversarial weight perturbation defense · CIFAR-100 | accuracy under attack (%): 28.86 -> 28.77 | Hurt |
| Use LAFEAT MT(1k,10) instead of AutoAttack AA100 | Overfitting defense · CIFAR-100 | accuracy under attack (%): 18.95 -> 18.86 | Hurt |
| Add adversarial training of latent logits layers (ablation) | PGD-7-trained WideResNet-32-10 · CIFAR-10 | clean accuracy (%): 79.56 -> 83.53 | Improved |
| Add adversarial training of latent logits layers (ablation) | PGD-7-trained WideResNet-32-10 · CIFAR-10, PGD-100 | accuracy under attack (%): 47.06 -> 47.17 | Improved |
| Add adversarial training of latent logits layers (ablation) | PGD-7-trained WideResNet-32-10 · CIFAR-10, LAFEAT --LF attack | accuracy under attack (%): 41.21 -> 41.50 | Improved |
| Add adversarial training of latent logits layers (ablation) | PGD-7-trained WideResNet-32-10 · CIFAR-10, LAFEAT +LF attack | accuracy under attack (%): 41.05 -> 41.26 | Improved |
| Add latent features to an existing attack-tactic combination (ablation) | LAFEAT · 14 defending models | accuracy under attack (%) | Improved |
| Vary the interpolation weight between latent and output logits (ablation) | LAFEAT · CIFAR-10 defense models | accuracy under attack (%) | Mixed |
| Train auxiliary logits using the attack dataset rather than the training dataset (ablation) | LAFEAT · Adversarial attack evaluation | attack strength | No clear effect |
Closest papers by meaning
- LAFIT: Efficient and Reliable Evaluation of Adversarial Defenses With Latent Features (TPAMI 2023 · similarity 0.95)
- Towards the first adversarially robust neural network model on MNIST (ICLR 2019 · similarity 0.89)
- Reliably Fast Adversarial Training via Latent Adversarial Perturbation (ICCV 2021 · similarity 0.87)
- Adversarial Robustness via Deformable Convolution with Stochasticity (ICML 2025 · similarity 0.86)
- SuperDeepFool: a new fast and accurate minimal adversarial attack (NeurIPS 2024 · similarity 0.86)
- Understanding the Robustness of Randomized Feature Defense Against Query-Based Adversarial Attacks (ICLR 2024 · 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.
