Research map / Adversarial robustness and security / Attacks on text and graphs
Adversarial Counterfactual Visual Explanations
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 |
|---|---|---|---|
| Apply DDPM filtering before the attack and refine the result with a localized inpainting mask | ACE · CelebA | FID: DiME 3.17; ACE ℓ1 1.27 | Helped |
| Apply DDPM filtering before the attack and refine the result with a localized inpainting mask | ACE · BDD100k | Flip Rate: DiME 90.5; ACE ℓ1 99.9 | Helped |
| Use ℓ1 rather than ℓ2 distance loss | ACE · CelebA Smile | MNAC: 2.94 vs 2.77 | Mixed |
| Use ℓ1 rather than ℓ2 distance loss | ACE · CelebA HQ Smile | MNAC: 1.56 vs 1.87 | Helped |
| Omit the RePaint refinement stage (ablation) | ACE pre-explanations · CelebA Smile | sFID: 4.63 -> 3.97 | Hurt |
| Use all diffusion steps instead of re-spaced steps (ablation) | ACE · CelebA Smile | Flip Rate: 99.97 -> 93.37 | Hurt |
| Use all diffusion steps instead of re-spaced steps (ablation) | ACE · CelebA Age | Flip Rate: 99.77 -> 93.50 | Hurt |
| Use refinement/inpainting after the pre-explanation (ablation) | ACE · CelebA Smile | CD: 3.05 -> 1.73 | Helped |
| Use C&W or standard gradient descent instead of PGD (ablation) | ACE · ACE attack optimization | optimization iterations: twice as many iterations | Hurt |
| Increase diversity by changing refinement re-spacing and sampled noise (ablation) | ACE · CelebA | LPIPS diversity σL: 0.110 -> 0.1436 | Helped |
| Use ACE rather than DiME | counterfactual explanations · CelebA HQ Age | sFID: ACE ℓ2 28.2; DiME 27.8 | Hurt |
Closest papers by meaning
- Promoting Counterfactual Robustness through Diversity (AAAI 2024 · similarity 0.88)
- Wide Two-Layer Networks can Learn from Adversarial Perturbations (NeurIPS 2024 · similarity 0.86)
- Adv-Diffusion: Imperceptible Adversarial Face Identity Attack via Latent Diffusion Model (AAAI 2024 · similarity 0.86)
- Constructing Semantics-Aware Adversarial Examples with a Probabilistic Perspective (NeurIPS 2024 · similarity 0.85)
- DiG-IN: Diffusion Guidance for Investigating Networks - Uncovering Classifier Differences Neuron Visualisations and Visual Counterfactual Explanations (CVPR 2024 · similarity 0.85)
- Theoretical Understanding of Learning from Adversarial Perturbations (ICLR 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.