Research map / Adversarial robustness and security / Black-box and transfer attacks
Generating Natural Adversarial Examples
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 |
|---|---|---|---|
| Use hybrid shrinking search instead of iterative stochastic search | Natural adversary generation · Latent-space adversary search | speedup: 4× speedup | Helped |
| Generate adversaries in the learned latent space rather than perturbing pixels directly | Black-box image classifiers on MNIST · MNIST | human preference for more natural changes: 78% | Helped |
| Use classifiers with higher test accuracy | MNIST classifiers · MNIST test | average Δz: 1.24 -> 1.61 | Mixed |
| Use classifiers with higher test accuracy | Textual Entailment classifiers · Textual Entailment test | average Δz: 0.12 -> 0.14 -> 0.26 | Mixed |
| Use classifiers with higher test accuracy | MNIST classifiers · MNIST test | P(largest Δz): 0.22 -> 0.78 | Helped |
| Use classifiers with higher test accuracy | Textual Entailment classifiers · Textual Entailment test | P(largest Δz): 0.15 -> 0.18 -> 0.66 | Helped |
| Use higher-accuracy classifiers | MNIST adversary judgments · MNIST pilot study | Which closer to original?: 0.87 vs 0.13 | Mixed |
| Use higher-accuracy classifiers | Textual Entailment adversary judgments · Textual Entailment pilot study | Is it similar to the original?: 0.81 vs 0.58 | Mixed |
Closest papers by meaning
- Constructing Unrestricted Adversarial Examples with Generative Models (NeurIPS 2018 · similarity 0.90)
- Generate (Non-Software) Bugs to Fool Classifiers (AAAI 2020 · similarity 0.90)
- Adversarial Examples Are Not Bugs, They Are Features (NeurIPS 2019 · similarity 0.88)
- GAT: Generative Adversarial Training for Adversarial Example Detection and Robust Classification (ICLR 2020 · similarity 0.88)
- One Man's Trash Is Another Man's Treasure: Resisting Adversarial Examples by Adversarial Examples (CVPR 2020 · similarity 0.88)
- Constructing Semantics-Aware Adversarial Examples with a Probabilistic Perspective (NeurIPS 2024 · similarity 0.88)
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.