Research map / Adversarial robustness and security / Certified and empirical robustness
Detecting Adversarial Samples Using Influence Functions and Nearest Neighbors
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 |
|---|---|---|---|
| Combine influence-selected training examples with k-NN ranks and distances | NNIF · CIFAR-10 / Deepfool | AUC (%): 99.32 vs 97.49 | Helped |
| Combine influence-selected training examples with k-NN ranks and distances | NNIF · CIFAR-100 / CW | AUC (%): 96.51 vs 74.43 | Helped |
| Combine influence-selected training examples with k-NN ranks and distances | NNIF · SVHN / EAD | AUC (%): 97.40 vs 95.77 | Helped |
| Combine influence-selected training examples with k-NN ranks and distances | NNIF · CIFAR-100 / EAD | AUC (%): 74.86 vs 78.42 | Hurt |
| Use only helpful-example distance (D^M↑) (ablation) | NNIF · CIFAR-10 / Deepfool | AUC (%): 82.11 -> 99.79 | Helped |
| Add harmful-example rank (R^M↓) to helpful-example distance (ablation) | NNIF · CIFAR-10 / Deepfool | AUC (%): 99.79 -> 99.79 | No clear effect |
| Add helpful-example rank (R^M↑) to helpful-example distance (ablation) | NNIF · CIFAR-10 / Deepfool | AUC (%): 99.79 -> 99.79 | No clear effect |
| Add harmful-example distance (D^M↓) to helpful-example distance (ablation) | NNIF · CIFAR-10 / Deepfool | AUC (%): 99.79 -> 99.82 | No clear effect |
| Use only harmful-example rank (R^M↓) (ablation) | NNIF · CIFAR-10 / Deepfool | AUC (%): 66.14 vs 99.82 | Hurt |
| Use only harmful-example distance (D^M↓) (ablation) | NNIF · CIFAR-10 / CW | AUC (%): 89.97 vs 82.11 | Mixed |
| Use only helpful-example rank (R^M↑) (ablation) | NNIF · CIFAR-10 / Deepfool and CW | AUC (%): 98.27 vs 81.91 | Mixed |
| Use only helpful-example distance (D^M↑) (ablation) | NNIF · CIFAR-10 / Deepfool and CW | AUC (%): 99.79 vs 97.27 | Mixed |
Closest papers by meaning
- Adversarial Examples for k-Nearest Neighbor Classifiers Based on Higher-Order Voronoi Diagrams (NeurIPS 2021 · similarity 0.89)
- Analyzing the Robustness of Nearest Neighbors to Adversarial Examples (ICML 2018 · similarity 0.88)
- Adversarial Training Can Provably Improve Robustness: Theoretical Analysis of Feature Learning Process Under Structured Data (ICLR 2025 · similarity 0.87)
- Theoretical Understanding of Learning from Adversarial Perturbations (ICLR 2024 · similarity 0.86)
- Wide Two-Layer Networks can Learn from Adversarial Perturbations (NeurIPS 2024 · similarity 0.86)
- Detecting Adversarial Data Using Perturbation Forgery (CVPR 2025 · 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.