Research map / Adversarial robustness and security / Attacks on text and graphs
ReTarget: Representation Transformation via Adversarial Regularization for Geometric Misalignment
Jia-You Chen, Shang-Tse 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 |
|---|---|---|---|
| use ReTarget instead of NoPeek | split inference defense · MSCOCO, DRAG, layer 3 | LPIPS: 0.4690 -> 0.6337 | Improved |
| use ReTarget instead of NoPeek | split inference defense · MSCOCO, DRAG, layer 3 | MS-SSIM: 0.3760 -> 0.0209 | Improved |
| use ReTarget instead of NoPeek | split inference defense · MSCOCO, DRAG, layer 12 | LPIPS: 0.7645 -> 0.8389 | Improved |
| add prototype-guided semantic loss (lambda 1 vs 0) (ablation) | ReTarget · ImageNet val, layer 3 | Disrupt@10: 0.697 -> 0.856 | Improved |
| add prototype-guided semantic loss (lambda 1 vs 0) (ablation) | ReTarget · ImageNet val, layer 12 | Disrupt@10: 0.501 -> 0.671 | Improved |
| allow adaptive attacker to train decoder on defended features (DRAG++) | ReTarget · MSCOCO, layer 3 | MS-SSIM: 0.0209 -> 0.0481 | Hurt |
| allow adaptive attacker (DRAG++) | ReTarget · MSCOCO, layer 9 | MS-SSIM: 0.2556 -> 0.1982 | Hurt |
| use ReTarget instead of NoPeek | split inference defense · FFHQ, GLASS attack | reconstruction similarity | Mixed |
| use ReTarget instead of NoPeek | split inference defense · ImageNet, layer 3 | Delta I(X;H): -0.394 -> -0.763 | Improved |
| use ReTarget instead of NoPeek | split inference defense · ImageNet, layer 12 | Delta I(X;H): -0.369 -> 0.019 | Hurt |
| use ReTarget instead of DISCO | split inference defense · MSCOCO, layer 3 | accuracy: 0.7207 -> 0.7461 | Improved |
Closest papers by meaning
- What Your Features Reveal: Data-Efficient Black-Box Feature Inversion Attack for Split DNNs (CVPR 2026 · similarity 0.84)
- Adversarial Reconstruction Feedback for Robust Fine-grained Generalization (ICCV 2025 · similarity 0.84)
- Generative Model Inversion Through the Lens of the Manifold Hypothesis (NeurIPS 2025 · similarity 0.84)
- GAN You See Me? Enhanced Data Reconstruction Attacks against Split Inference (NeurIPS 2023 · similarity 0.84)
- Sparse Model Inversion: Efficient Inversion of Vision Transformers for Data-Free Applications (ICML 2024 · similarity 0.84)
- Preserving Deep Representations in One-Shot Pruning: A Hessian-Free Second-Order Optimization Framework (ICLR 2025 · similarity 0.83)
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.
