Research map / Adversarial robustness and security / Attacks on text and graphs
Manifold Projection for Adversarial Defense on Face Recognition
Jianli Zhou, Chao Liang, Jun Chen
Read the paperCheck 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 |
|---|---|---|---|
| add A-VAE input purification | ResNet-50 face verifier · LFW grey-box FGSM eps=8 | verification accuracy (average): 0.245 -> 0.753 | Improved |
| A-VAE instead of adversarial training | ResNet-50 face verifier · LFW grey-box FGSM eps=8 | verification accuracy (average): 0.457 -> 0.753 | Improved |
| add A-VAE input purification | ResNet-50 face verifier · LFW grey-box PGD eps=8 | verification accuracy (average): 0.003 -> 0.828 | Improved |
| add A-VAE input purification | ResNet-50 face verifier · LFW grey-box C&W | verification accuracy (average): 0.008 -> 0.610 | Improved |
| add A-VAE input purification | ResNet-50 face verifier · LFW clean images | verification accuracy (average): 0.992 -> 0.963 | Hurt |
| attacker knows defense (white-box via BPDA) instead of grey-box | A-VAE · LFW PGD eps=8 | verification accuracy (average): 0.828 -> 0.573 | Hurt |
| attacker knows defense (white-box via BPDA) instead of grey-box | A-VAE · LFW FGSM eps=8 | verification accuracy (average): 0.753 -> 0.552 | Hurt |
| add A-VAE input purification | ResNet-50 face verifier · LFW on-manifold attack | verification accuracy (average): 0.321 -> 0.667 | Improved |
| use A-VAE rather than HGD | ResNet-50 face verifier · LFW white-box PGD eps=8 | verification accuracy (average): 0.518 -> 0.573 | Improved |
| increase generator input size from 32x32 to 128x128 (ablation) | A-VAE · LFW grey-box FGSM eps=8 | robustness vs reconstruction quality | Mixed |
| replace pixel-wise reconstruction loss with adversarial loss (ablation) | A-VAE · LFW | robustness | Improved |
| downsample input to 32x32 before encoder (ablation) | A-VAE · LFW | robustness and fidelity | Improved |
Earlier methods this paper compares with
Named baselines from earlier accepted papers in this paper's results tables: better than 1 of 1 on most shared results.
- ComDefend — ComDefend: An Efficient Image Compression Model to Defend Adversarial Examples (CVPR 2019) · better on 3 of 3 shared results
Matched by method name in results tables read by AI. Authors' own tables tend to favour the newer method, and a comparison can be missed when a method is renamed.
Closest papers by meaning
- Dual Manifold Adversarial Robustness: Defense against Lp and non-Lp Adversarial Attacks (NeurIPS 2020 · similarity 0.91)
- Enhancing the Adversarial Robustness via Manifold Projection (AAAI 2025 · similarity 0.89)
- Discrete Point-Wise Attack Is Not Enough: Generalized Manifold Adversarial Attack for Face Recognition (CVPR 2023 · similarity 0.88)
- Adversarial Purification with the Manifold Hypothesis (AAAI 2024 · similarity 0.88)
- Explaining the role of Intrinsic Dimensionality in Adversarial Training (ICML 2025 · similarity 0.86)
- Manifold Constraints for Imperceptible Adversarial Attacks on Point Clouds (AAAI 2024 · similarity 0.85)
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.
