Research map / Adversarial robustness and security / Black-box and transfer attacks
AdvFlow: Inconspicuous Black-box Adversarial Attacks using Normalizing Flows
Hadi Mohaghegh Dolatabadi, Sarah Erfani, Christopher Leckie
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 a clean-data-pretrained normalizing flow instead of N-Attack's search distribution | black-box attack · CIFAR-10 defended classifiers | attack success rate: 38.97 -> 41.21; 36.90 -> 40.22; 38.04 -> 40.67 | Improved |
| Use a clean-data-pretrained normalizing flow instead of N-Attack's search distribution | black-box attack · SVHN defended classifiers | attack success rate: 50.28 -> 50.76; 35.42 -> 41.49; 41.49 -> 44.22 | Improved |
| Use AdvFlow rather than N-Attack | black-box attack · CIFAR-10 vanilla classifier | attack success rate: 100 -> 99.42 | Hurt |
| Use AdvFlow rather than N-Attack | black-box attack · SVHN vanilla classifier | attack success rate: 98.76 -> 90.31 | Hurt |
| Pretrain the flow on clean data rather than use random flow weights (ablation) | AdvFlow · CIFAR-10, LID detector | detection AUROC: 84.39 -> 57.59 | Improved |
| Pretrain the flow on clean data rather than use random flow weights (ablation) | AdvFlow · CIFAR-10, Mahalanobis detector | detection AUROC: 99.50 -> 66.85 | Improved |
| Pretrain the flow on clean data rather than use random flow weights (ablation) | AdvFlow · CIFAR-10, Res-Flow detector | detection AUROC: 99.40 -> 67.03 | Improved |
| Pretrain the flow on clean data rather than use random flow weights (ablation) | AdvFlow · SVHN, LID detector | detection AUROC: 58.92 -> 61.11 | Hurt |
| Pretrain the flow on clean data rather than use random flow weights (ablation) | AdvFlow · SVHN, Mahalanobis detector | detection AUROC: 74.67 -> 64.72 | Improved |
| Pretrain the flow on clean data rather than use random flow weights (ablation) | AdvFlow · SVHN, Res-Flow detector | detection AUROC: 74.86 -> 64.68 | Improved |
| Use AdvFlow rather than N-Attack | black-box attack · CIFAR-10 defended classifiers | average query count: 874.91 -> 421.63; 973.05 -> 436.8; 941.67 -> 424.95 | Mixed |
| Use AdvFlow rather than N-Attack | black-box attack · SVHN defended classifiers | average query count: 903.12 -> 692.73; 1243.38 -> 526.37; 756.48 -> 480.02 | Mixed |
Later papers that compared with this method
1 later accepted paper reports AdvFlow as a baseline in their results tables; 1 of them report a better number on most of the results they share with it.
- Generalizable Black-Box Adversarial Attack With Meta Learning (TPAMI 2023 · better on 1 of 1 shared result)
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.
- N-Attack — NATTACK: Learning the Distributions of Adversarial Examples for an Improved Black-Box Attack on Deep Neural Networks (ICML 2019) · better on 11 of 14 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
- Transferable Adversarial Perturbations (ECCV 2018 · similarity 0.89)
- GeoDA: A Geometric Framework for Black-Box Adversarial Attacks (CVPR 2020 · similarity 0.87)
- BlackboxBench: A Comprehensive Benchmark of Black-Box Adversarial Attacks (TPAMI 2025 · similarity 0.87)
- Simple Black-box Adversarial Attacks (ICML 2019 · similarity 0.87)
- Black-box Adversarial Attacks with Limited Queries and Information (ICML 2018 · similarity 0.87)
- Strong Transferable Adversarial Attacks via Ensembled Asymptotically Normal Distribution Learning (CVPR 2024 · similarity 0.86)
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.
