Research map / Adversarial robustness and security / Black-box and transfer attacks
Towards Reverse-Engineering Black-Box Neural Networks
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 |
|---|---|---|---|
| Optimize query inputs and add an output interpreter (kennen-io) rather than use fixed queries with kennen-o | MNIST-NETS attribute prediction · MNIST-NETS Random split | mean attribute prediction accuracy: 73.4 -> 80.1 | Helped |
| Use ranking outputs instead of probability outputs (ablation) | kennen-o · MNIST-NETS Random split | mean attribute prediction accuracy: 73.4 -> 69.7 | Hurt |
| Use bottom-1 labels instead of probability outputs (ablation) | kennen-o · MNIST-NETS Random split | mean attribute prediction accuracy: 73.4 -> 54.4 | Hurt |
| Use top-1 labels instead of probability outputs (ablation) | kennen-o · MNIST-NETS Random split | mean attribute prediction accuracy: 73.4 -> 39.5 | Hurt |
| Use one crafted query instead of 100 fixed queries (ablation) | MNIST-NETS attribute prediction · MNIST-NETS Random split | mean attribute prediction accuracy: 73.4 -> 52.7 | Hurt |
| Increase the number of queries from 100 to 1,000 (ablation) | kennen-o · ImageNet classifier family prediction | accuracy: 90.4 -> 94.8 | Helped |
| Predict the target family and generate AIPs against that family instead of multiple families | ImageNet adversarial attack · ImageNet adversarial attack | misclassification rate: 82.2 -> 85.7 | Helped |
| Generate AIPs within the target architecture family rather than across families | GAMAN · ImageNet transferability | misclassification rate: ResNet target: 87 vs 78 | Mixed |
Closest papers by meaning
- Black-box Certification and Learning under Adversarial Perturbations (ICML 2020 · similarity 0.89)
- Understanding the Robustness of Randomized Feature Defense Against Query-Based Adversarial Attacks (ICLR 2024 · similarity 0.88)
- Black-box Adversarial Attacks with Limited Queries and Information (ICML 2018 · similarity 0.88)
- Black-Box Adversarial Attack with Transferable Model-based Embedding (ICLR 2020 · similarity 0.87)
- NATTACK: Learning the Distributions of Adversarial Examples for an Improved Black-Box Attack on Deep Neural Networks (ICML 2019 · similarity 0.87)
- Towards Lightweight Black-Box Attack Against Deep Neural Networks (NeurIPS 2022 · 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.